Base_JSON/prefixF/json/fnp/2020.fnp-1.36.json

{
    "paper_id": "2020",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T10:23:51.517252Z"
    },
    "title": "Extracting Fine-Grained Economic Events from Business News",
    "authors": [
        {
            "first": "Gilles",
            "middle": [],
            "last": "Jacobs",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "V\u00e9ronique Hoste Language and Translation Technology Team Ghent University",
                "location": {
                    "postCode": "9000",
                    "settlement": "Gent",
                    "country": "Belgium"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Based on a recently developed fine-grained event extraction dataset for the economic domain, we present in a pilot study for supervised economic event extraction. We investigate how a stateof-the-art model for event extraction performs on the trigger and argument identification and classification. While F 1-scores of above 50% are obtained on the task of trigger identification, we observe a large gap in performance compared to results on the benchmark ACE05 dataset. We show that single-token triggers do not provide sufficient discriminative information for a finegrained event detection setup in a closed domain such as economics, since many classes have a large degree of lexico-semantic and contextual overlap.",
    "pdf_parse": {
        "paper_id": "2020",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Based on a recently developed fine-grained event extraction dataset for the economic domain, we present in a pilot study for supervised economic event extraction. We investigate how a stateof-the-art model for event extraction performs on the trigger and argument identification and classification. While F 1-scores of above 50% are obtained on the task of trigger identification, we observe a large gap in performance compared to results on the benchmark ACE05 dataset. We show that single-token triggers do not provide sufficient discriminative information for a finegrained event detection setup in a closed domain such as economics, since many classes have a large degree of lexico-semantic and contextual overlap.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "We present a pilot study on a novel dataset annotated with economic and financial events in English company-specific news. Event processing automatically obtains the \"what, who, where and when\" of real-world events described in text. Event extraction consists of identifying event triggers, i.e. the tokens that express an event of a predetermined type, and identifying participant arguments, i.e. the tokens that express prototypical participant roles.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Event extraction is typically an upstream step in pipelines for financial applications: it has been used for news summarization of single (Lee et al., 2003; Marujo et al., 2017) or multiple documents (Liu et al., 2007; Glava\u0161 and\u0160najder, 2014) , forecasting and market analysis (Nassirtoussi et al., 2014; Bholat et al., 2015; Nardo et al., 2016; Zhang et al., 2018; Chen et al., 2019) , risk analysis (Hogenboom et al., 2015; , policy assessment (Tobback et al., 2018; Karami et al., 2018) , and marketing (Rambocas and Pacheco, 2018) . This work aims to enable these information extraction tasks in the financial domain by making available a dataset and fine-grained event extraction model for companyspecific news by classifying economic event triggers and participant arguments in text.",
                "cite_spans": [
                    {
                        "start": 138,
                        "end": 156,
                        "text": "(Lee et al., 2003;",
                        "ref_id": "BIBREF27"
                    },
                    {
                        "start": 157,
                        "end": 177,
                        "text": "Marujo et al., 2017)",
                        "ref_id": "BIBREF34"
                    },
                    {
                        "start": 200,
                        "end": 218,
                        "text": "(Liu et al., 2007;",
                        "ref_id": "BIBREF32"
                    },
                    {
                        "start": 219,
                        "end": 243,
                        "text": "Glava\u0161 and\u0160najder, 2014)",
                        "ref_id": null
                    },
                    {
                        "start": 278,
                        "end": 305,
                        "text": "(Nassirtoussi et al., 2014;",
                        "ref_id": "BIBREF36"
                    },
                    {
                        "start": 306,
                        "end": 326,
                        "text": "Bholat et al., 2015;",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 327,
                        "end": 346,
                        "text": "Nardo et al., 2016;",
                        "ref_id": "BIBREF35"
                    },
                    {
                        "start": 347,
                        "end": 366,
                        "text": "Zhang et al., 2018;",
                        "ref_id": "BIBREF49"
                    },
                    {
                        "start": 367,
                        "end": 385,
                        "text": "Chen et al., 2019)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 402,
                        "end": 426,
                        "text": "(Hogenboom et al., 2015;",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 447,
                        "end": 469,
                        "text": "(Tobback et al., 2018;",
                        "ref_id": "BIBREF42"
                    },
                    {
                        "start": 470,
                        "end": 490,
                        "text": "Karami et al., 2018)",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 507,
                        "end": 535,
                        "text": "(Rambocas and Pacheco, 2018)",
                        "ref_id": "BIBREF39"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Our SENTiVENT dataset of company-specific events was conceived to be compatible with finegrained event representations of the ACE benchmark corpora as to enable direct application of advances in the field. In our pilot study, we investigate the portability of an existing state-of-the-art model for event extraction, named DYGIE++ (Wadden et al., 2019b) , to the task of financial event extraction.",
                "cite_spans": [
                    {
                        "start": 331,
                        "end": 353,
                        "text": "(Wadden et al., 2019b)",
                        "ref_id": "BIBREF45"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Our work on economic event extraction is accommodated within the rich history of automatic event detection. The ACE (Automatic Content Extraction) annotation scheme and programme was highly influential in event processing. Periodically, \"Event Detection and Recognition\" evaluation competitions were organized where event extraction corpora were released (Consortium, 2005; Walker et al., 2006) to enable automatic inference of entities mentioned in text, the relations among entities, and the events in which these entities participate (Doddington et al., 2004) . Some years later, the ERE (Entities, Relations, Events) standard was conceived as a continuation of ACE with the goal of improving annotation consistency and quality. Our annotation scheme is inspired by the Rich ERE Event annotation schemes (Linguistic Data Consortium, 2016; Linguistic Data Consortium, 2015a; Linguistic Data Consortium, 2015b) as to provide compatibility with on-going research in event extraction, where the ACE and ERE datasets remain dominant benchmarks.",
                "cite_spans": [
                    {
                        "start": 355,
                        "end": 373,
                        "text": "(Consortium, 2005;",
                        "ref_id": null
                    },
                    {
                        "start": 374,
                        "end": 394,
                        "text": "Walker et al., 2006)",
                        "ref_id": null
                    },
                    {
                        "start": 537,
                        "end": 562,
                        "text": "(Doddington et al., 2004)",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 807,
                        "end": 841,
                        "text": "(Linguistic Data Consortium, 2016;",
                        "ref_id": null
                    },
                    {
                        "start": 842,
                        "end": 876,
                        "text": "Linguistic Data Consortium, 2015a;",
                        "ref_id": null
                    },
                    {
                        "start": 877,
                        "end": 911,
                        "text": "Linguistic Data Consortium, 2015b)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "Many approaches to the detection of economic events are knowledge-based and pattern-based (Arendarenko and Kakkonen, 2012; Hogenboom et al., 2013; Du et al., 2016; Chen et al., 2019) . These use rule-sets or ontology knowledge-bases which are largely or fully created by hand and do not rely fully on manually annotated supervised datasets for machine learning. The Stock Sonar project (Feldman et al., 2011a) notably uses domain experts to formulate event rules for rule-based stock sentiment analysis. Their approach has been successful in formulating trading strategies (Ben Ami and Feldman, 2017) and in assessing the impact of events on the stock market (Boudoukh et al., 2016) . Along the same line, Hogenboom et al. (2013) rely on a hand-crafted financial event ontology for pattern-based event detection in the economic domain and incorporates lexicons, gazetteers, PoS-tagging and morphological analysis.",
                "cite_spans": [
                    {
                        "start": 107,
                        "end": 122,
                        "text": "Kakkonen, 2012;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 123,
                        "end": 146,
                        "text": "Hogenboom et al., 2013;",
                        "ref_id": "BIBREF17"
                    },
                    {
                        "start": 147,
                        "end": 163,
                        "text": "Du et al., 2016;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 164,
                        "end": 182,
                        "text": "Chen et al., 2019)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 386,
                        "end": 409,
                        "text": "(Feldman et al., 2011a)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 573,
                        "end": 600,
                        "text": "(Ben Ami and Feldman, 2017)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 659,
                        "end": 682,
                        "text": "(Boudoukh et al., 2016)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 706,
                        "end": 729,
                        "text": "Hogenboom et al. (2013)",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "Several semi-or distantly supervision approaches exist in which seed sets are manually labeled or rule-sets are used to generate or enhance training data (Qian et al., 2019; R\u00f6nnqvist and Sarlin, 2017; Ein-Dor et al., 2019) . For Chinese, Yang et al. (2018) and Chen et al. (2019) rely on a knowledge-base of rules for extracting ACE-like events in stock market prediction. Han et al. (2018) describe a hybrid approach to ACE-like event extraction by labeling triggers, argument types, and event types for Chinese news articles with 8 economic event types using an automatically expanded trigger dictionary and used a pattern-matching approach for argument extraction.",
                "cite_spans": [
                    {
                        "start": 154,
                        "end": 173,
                        "text": "(Qian et al., 2019;",
                        "ref_id": "BIBREF38"
                    },
                    {
                        "start": 174,
                        "end": 201,
                        "text": "R\u00f6nnqvist and Sarlin, 2017;",
                        "ref_id": "BIBREF40"
                    },
                    {
                        "start": 202,
                        "end": 223,
                        "text": "Ein-Dor et al., 2019)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 239,
                        "end": 257,
                        "text": "Yang et al. (2018)",
                        "ref_id": "BIBREF48"
                    },
                    {
                        "start": 262,
                        "end": 280,
                        "text": "Chen et al. (2019)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 374,
                        "end": 391,
                        "text": "Han et al. (2018)",
                        "ref_id": "BIBREF16"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "Few strictly supervised approaches exist due to the lack of human-annotated ground-truth data. Malik et al. 2011annotated Dividend and Profit figure slots and detected these types with high precision by combining supervised learning with rule post-processing. The English and Dutch SentiFM business news corpus (Van De Kauter et al., 2015) contains token-span annotations of 10 event types with only one type of relation. Sentence-level event detection experiments have been conducted (Lefever and Hoste, 2016; Jacobs et al., 2018) , but supervised fine-grained extraction of triggers and roles as presented here. We are not aware of any other published fine-grained ACE-like event extraction approaches for the economic domain.",
                "cite_spans": [
                    {
                        "start": 485,
                        "end": 510,
                        "text": "(Lefever and Hoste, 2016;",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 511,
                        "end": 531,
                        "text": "Jacobs et al., 2018)",
                        "ref_id": "BIBREF22"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "(1) Moeller was named CFO in January 2009 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "EMPLOYMENT | START Employee Title",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Research",
                "sec_num": "2"
            },
            {
                "text": "(2) American Airlines reports load factor increase in May , shares gain",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Time",
                "sec_num": null
            },
            {
                "text": "REVENUE | INCREASE Company Time SECURITYVALUE | INCREASE",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Time",
                "sec_num": null
            },
            {
                "text": "(3) AbbVie 's dividend yield currently stands at 3.9% . ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Security",
                "sec_num": null
            },
            {
                "text": "We define economic events as 'textually reported real-world occurrences, actions, relations, and situations involving companies and firms'. Fine-grained events in our dataset are operationalized as event triggers, i.e. the minimal span of text (a single word or a phrase) that most succinctly expresses the occurrence of an event type, and event arguments, i.e., participating entities involved in the event by filling a certain prototypical semantic role. Figure 1 shows examples of annotated event triggers with their TYPE | SUBTYPE label linked to argument with their roles. We annotated 288 documents with structured events containing event types and subtypes, arguments and event attributes (i.e., negation, modality & event co-reference). The news article pertain to 30 S&P500 companies spread over different industries and time between June 2016 to May 2017 to avoid topical specialization. These companies were selected as a starting point for scraping because market data (e.g., key financials, spot prices, dividend, earnings and P/E ratio) for these are readily available, enabling market research. For a full description of the data collection, annotation process, dataset properties, and agreement study, we refer to Jacobs and Hoste (2020). The event typology contains 18 event main types with 42 subtypes for which 48 possible argument roles exist. Not every main event type has subtypes and the arguments \"Time\", \"Place\", and \"Capital\" can belong to any event. The typology was iteratively developed on a sample corpus in which news events were evaluated by financial domain experts. Starting with the set of event types in previous literature (Feldman et al., 2011b; Boudoukh et al., 2019; Hogenboom et al., 2013; Van De Kauter et al., 2015; Du et al., 2016) , types were removed and added based on frequency and cohesion of categories. Relevant argument roles were also added in this process of iterative refinement. For a full description of the types and subtypes, we refer to the annotation guidelines (Jacobs, 2020b) . Here is a list of main Types \u2192 unique Arguments:",
                "cite_spans": [
                    {
                        "start": 1660,
                        "end": 1683,
                        "text": "(Feldman et al., 2011b;",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 1684,
                        "end": 1706,
                        "text": "Boudoukh et al., 2019;",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 1707,
                        "end": 1730,
                        "text": "Hogenboom et al., 2013;",
                        "ref_id": "BIBREF17"
                    },
                    {
                        "start": 1731,
                        "end": 1758,
                        "text": "Van De Kauter et al., 2015;",
                        "ref_id": "BIBREF43"
                    },
                    {
                        "start": 1759,
                        "end": 1775,
                        "text": "Du et al., 2016)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 2023,
                        "end": 2038,
                        "text": "(Jacobs, 2020b)",
                        "ref_id": "BIBREF24"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 457,
                        "end": 465,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "3"
            },
            {
                "text": "\u2022 CSR/BRAND \u2192 Company \u2022 DEAL \u2192 Goal, Partner \u2022 DIVIDEND \u2192 Amount, Company, Historica- lYieldRatio, YieldRatio \u2022 EMPLOYMENT \u2192 Amount, Employee, Em- ployer, Replacer, Replacing, Title \u2022 EXPENSE \u2192 Amount, Company, Histori- calAmount \u2022 FACILITY \u2192 Company, Facility \u2022 FINANCIALREPORT \u2192 Reportee, Result \u2022 FINANCING \u2192 Amount, Financee, Financer \u2022 INVESTMENT \u2192 CapitalInvested, Investee, In- vestor \u2022 LEGAL \u2192 Allegation, Adjudicator, Com- plainant, Defendant, Sentence \u2022 MACROECONOMICS \u2192 AffectedCompany, Sector \u2022 MERGER/ACQUISITION \u2192 Acquirer, Cost, Target \u2022 PRODUCT/SERVICE \u2192 Producer, Prod- uct/Service, Trialer \u2022 PROFIT/LOSS \u2192 Amount, HistoricalAmount, Profiteer \u2022 RATING \u2192 Analyst, HistoricalRating, Secu- rity, TargetPrice \u2022 REVENUE \u2192 Amount, Company, De- creaseAmount, HistoricalAmount, In- creaseAmount \u2022 SALESVOLUME \u2192 Amount, Buyer, GoodsService, HistoricalAmount, Seller \u2022 SECURITYVALUE \u2192 DecreaseAmount, His- toricalPrice, IncreaseAmount, Price, Security",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "3"
            },
            {
                "text": "The types PROFIT/LOSS, REVENUE, EXPENSE, and SALESVOLUME represent major metrics in financial reporting on the income statement. Of all financial metrics, we found these types to be commonly and generally reported in business news. Other metrics such as debt ratios or cash flow statements as well as highly industry-specific indicators were collected under the FINANCIALREPORT type. Discussion of asset performance are captured by RATING and SECURITYVALUE. MACROECONOMICS is a broad category pertaining to events that do not involve decisions or interactions of specific companies. While this category is not company-specific, it was devised to capture discussions of sector trends, economywide phenomena, and governmental policy. In the event extraction pilot study, we only experiment with classifying main event types and not the subtypes. Table 1 shows the counts of annotation units in our SENTiVENT corpus compared to the benchmark ACE2005-English-Events corpus (henceforth ACE05) (Walker et al., 2006) for the training, development and test set splits used in the pilot experiments.",
                "cite_spans": [
                    {
                        "start": 988,
                        "end": 1009,
                        "text": "(Walker et al., 2006)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 844,
                        "end": 851,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "3"
            },
            {
                "text": "An important difference in event trigger annotation with ACE05 is the tagging of discontinuous, multiword triggers (e.g., \"upgraded ... to buy\", \"cut back ... expenses\", \"EPS decline\"). In ACE05, triggers are always single-word and any multi-word idiomatic expressions are joined into one token (e.g. \"kick SENTiVENT ACE05  Train  Dev  Test  Total  Train  Dev  Test  Total  Documents  228  30  30  288  529  28  40  597  Events  4,603  618  982  6,203  4,202  450  403  5,055  Sentences  5,475  681  727  6,883 17,172  923  832 18,927  Arguments  10,052  1,327  2,296  13, the bucket\" becomes \"kick the bucket\"). For SENTiVENT, 42% of triggers are multi-word and 13% are both discontinuous and multi-word.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 317,
                        "end": 572,
                        "text": "ACE05  Train  Dev  Test  Total  Train  Dev  Test  Total  Documents  228  30  30  288  529  28  40  597  Events  4,603  618  982  6,203  4,202  450  403  5,055  Sentences  5,475  681  727  6,883 17,172  923  832 18,927  Arguments  10,052  1,327  2,296  13,",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "3"
            },
            {
                "text": "Comparing annotation counts in ACE05 to our annotations, SENTiVENT has less documents but a higher density of event annotations. This reflects a difference in annotation approach in which events are less constrained by syntactic rules than in ACE05. Unlike ACE, we did not annotate entity tags and do not restrict event arguments definition by entity type. Not restricting events by syntactic rules or entity type results in a conceptually simplified annotation process. It also decreases to degree to which lexicosemantically present events are not tagged because of restrictive rules. Furthermore, the generality of geo-political event types in ACE05 compared to the domain-specific economic events in business news entails that more relevant events will be present. Due to the absence of entity annotations in our dataset, we also have significantly more arguments as in ACE05 argument roles are restricted by specific entity labels. Our argument role annotation is any span within the event scope that expresses the argument role and is thus semantically motivated.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Dataset",
                "sec_num": "3"
            },
            {
                "text": "Task definition Event extraction concerns identifying event triggers and their event types, event arguments and event roles. The input documents are represented as a sequence of tokens D from which the model constructs the set of all possible within-sentence spans S = {s 1 , ..., s T } in the document (limited by a threshold-length). Each token d i is assigned an event type label t i . Then, for each trigger d i , event arguments are assigned by predicting and argument role a i j for all spans s j in the same sentence as d i .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "Evaluation For ACE05, we follow the evaluation splits (cf. Table 1 ) and method commonly used in previous research (Nguyen et al., 2016; Sha et al., 2018; Zhang et al., 2019) . All reported precision (P), recall (R), and F 1 -scores (F 1 ) are micro-averaged. Our experiments involve the following four subtasks:",
                "cite_spans": [
                    {
                        "start": 115,
                        "end": 136,
                        "text": "(Nguyen et al., 2016;",
                        "ref_id": "BIBREF37"
                    },
                    {
                        "start": 137,
                        "end": 154,
                        "text": "Sha et al., 2018;",
                        "ref_id": "BIBREF41"
                    },
                    {
                        "start": 155,
                        "end": 174,
                        "text": "Zhang et al., 2019)",
                        "ref_id": "BIBREF50"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 59,
                        "end": 66,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "Trigger identification (Trig-ID) is the subtask of identifying if a token position matches a ground-truth reference trigger. Trigger classification (Trig-C) determines the type of the identified trigger. A trigger is correct when it is correctly identified and its type label matches the reference.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "Argument identification (Arg-ID) is the subtask of identifying if a text span belongs to a certain event type. An argument is correctly identified when the span matches and the event type is correct. Argument classification (Arg-C) determines the role (e.g. \"Employee\") of an identified argument. An argument is correctly classified if it is correctly identified and its role label matches the ground-truth reference.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "Data pre-processing All documents were sentence split and tokenized before annotation. As our event triggers are multi-token spans similar to Rich-ERE triggers, we transformed these multi-token spans into single-token triggers to allow us to use the DYGIE++ implementation in its basic configuration. For argument spans this pre-processing step is not needed as arguments are annotated as continuous, multitoken spans in both ACE05 and SENTiVENT. We selected single tokens by dependency parsing the triggers and selecting verbal and nominal root token within the trigger span using Spacy NLP library (Honnibal and Montani, 2017) . When filtering multiple tokens to a single trigger, we use syntactic priority criteria based on PoS in which nominal tokens are prefered over verbal tokens which are prefered over other types. Search for these parts-of-speech goes one level down into the parse tree from the original trigger's root token, defaulting to the original root token if no nominal of verbal child is found. Examples of this pre-processing applied to original multi-token triggers:",
                "cite_spans": [
                    {
                        "start": 600,
                        "end": 628,
                        "text": "(Honnibal and Montani, 2017)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "\u2022 PROFIT/LOSS \"improve margins\" \u2192 \"margins\" \u2022 FACILITY \"operating on-site gas stations\" \u2192 \"stations\" \u2022 SALESVOLUME \"attracting the greatest volume\" \u2192 \"volume\" This priority chain was empirically determined on a subsample of 15% of all triggers by comparing other selection criteria.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "The DYGIE++ model also allows the use of named entity labels as features in training the trigger and argument subtasks. Unlike ACE05, our dataset does not include entity annotations (argument roles are unconstrained by entity types). To test if silver-standard entity labels improve identification and classification, we used entity label predictions of a trained entity model on the ACE05 dataset (Entity F 1 -score: 90.7%).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "Event extraction model For the event extraction pipeline we relied on the state-of-the-art DYGIE++ information extraction framework (Wadden et al., 2019b; Wadden et al., 2019a) . As a current stateof-the-art model on ACE event extraction and other information extraction tasks, DYGIE++ is a good candidate for establishing a baseline on our new dataset. DYGIE++ relies on graph propagation of contextualized embeddings of text spans in the local and global context to jointly model event triggers and arguments. The model enumerates candidate text spans and encodes them using pretrained contextual language models (such as BERT (Devlin et al., 2019) ) and task-specific message updates passed over a text span graph. BERT embeddings are encoded at the subword-level by the WordPiece algorithm (Kudo, 2018) and to obtain token-level representations WordPieces are pooled. Tokens are encoded with frozen BERT embeddings using a sliding window, feeding each sentence to BERT together with a size-L window of surrounding sentences. Spans of text are enumerated and constructed by concatenating the tokens together with a learned span width embedding. With event propagation enabled, a graph is generated based on the model's best guess at the relations present among spans in the document. The event graph has two types of nodes: trigger nodes and argument nodes. Trigger nodes pass messages to their likely arguments updating those argument representations and arguments pass messages to their probable triggers. The re-contextualized representations are input into a two-layer feedforward neural net (FFNN) for each subtask. For a trigger token g i , the final prediction is computed by FFNN trig (g i ) and for argument role prediction the relevant pair of embeddings is concatenated and FFNN arg ([g i , g j ]) computed.",
                "cite_spans": [
                    {
                        "start": 132,
                        "end": 154,
                        "text": "(Wadden et al., 2019b;",
                        "ref_id": "BIBREF45"
                    },
                    {
                        "start": 155,
                        "end": 176,
                        "text": "Wadden et al., 2019a)",
                        "ref_id": "BIBREF44"
                    },
                    {
                        "start": 629,
                        "end": 650,
                        "text": "(Devlin et al., 2019)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 794,
                        "end": 806,
                        "text": "(Kudo, 2018)",
                        "ref_id": "BIBREF26"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experimental Setup",
                "sec_num": "4"
            },
            {
                "text": "We experiment with two main variations in the model architecture: BERT+LSTM feeds pretrained BERT embeddings to a bi-directional LSTM layer, while the LSTM parameters are trained jointly with a task specific layer. BERT Finetune uses supervised fine-tuning of the BERT model on the end-task. We relied on the BERT implementations provided by the AllenNLP framework (Gardner et al., 2018) : BERT Base with 12 layers (transformer blocks), 12 attention heads, and 110 million parameters. and BERT Large with 24 layers, 16 attention heads and, 340 million parameters. In initial testing, we found BERT Large improved performance in the LSTM variant over BERT Base . For finetuning, we relied on the smaller BERT Base due to memory restrictions.",
                "cite_spans": [
                    {
                        "start": 365,
                        "end": 387,
                        "text": "(Gardner et al., 2018)",
                        "ref_id": "BIBREF14"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Model variations",
                "sec_num": null
            },
            {
                "text": "We examine the impact of disabling the NER subtask in the DYGIE++ event extraction pipeline. The SENTiVENT dataset does not contain manually annotated ground-truth entity labels. The labels used in experiments with joint training on the NER subtask are predictions from a pretrained ACE05 model. We also enable event propagation where graph updates between triggers and arguments are computed. The base approach (BERT+LSTM) with event propagation bypasses trigger-argument span updates and directly feeds the embedded span representations from the Bi-LSTM layer to the feedforward scoring layer.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Model variations",
                "sec_num": null
            },
            {
                "text": "Additionally, we experimented with replacing the pretrained BERT general language model with the in-domain FinBERT model (Araci, 2019) as previous state-of-the-art systems have shown in-domain pretraining to increase performance on text classification and information extraction tasks (Howard and Ruder, 2018; Wadden et al., 2019b) . FinBERT further pretrains BERT on the TRC2-financial corpus of news articles published by Reuters between 2008 and 2010 resulting in a model that outperforms regular BERT on financial sentiment analysis tasks. 1 Table 2 : Micro-avg. precision (P), recall (R), and F 1 -score of the best system (BERT-Large + LSTM) on SENTiVENT and ACE05 for comparison. Table 2 shows the results for the best configuration of the SENTiVENT model. The best model configuration was selected by the highest mean F 1 -score on the Trig-C and Arg-C subtasks. The scores for the best model variant with similar configuration trained on ACE05 is given as comparison. The best model variation on the different SENTiVENT event subtasks is BERT + LSTM with the NER subtask enabled ( Table 3) . The same architecture also was best for ACE05. 2 For our task, using the in-domain FinBERT did not improve performance over BERT on any task. In line with (Wadden et al., 2019b) , fine-tuning decreases performance for both BERT and FinBERT configurations. This is likely due to the sensitivity of both the optimization hyper-parameters and BERT finetuning as the trigger detector begins overfitting before the argument detector is finished training. Event propagation also does not improve scores, likely due to the asymmetric relationship between triggers and arguments.",
                "cite_spans": [
                    {
                        "start": 121,
                        "end": 134,
                        "text": "(Araci, 2019)",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 285,
                        "end": 309,
                        "text": "(Howard and Ruder, 2018;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 310,
                        "end": 331,
                        "text": "Wadden et al., 2019b)",
                        "ref_id": "BIBREF45"
                    },
                    {
                        "start": 1148,
                        "end": 1149,
                        "text": "2",
                        "ref_id": null
                    },
                    {
                        "start": 1256,
                        "end": 1278,
                        "text": "(Wadden et al., 2019b)",
                        "ref_id": "BIBREF45"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 546,
                        "end": 553,
                        "text": "Table 2",
                        "ref_id": null
                    },
                    {
                        "start": 687,
                        "end": 694,
                        "text": "Table 2",
                        "ref_id": null
                    },
                    {
                        "start": 1090,
                        "end": 1098,
                        "text": "Table 3)",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Model variations",
                "sec_num": null
            },
            {
                "text": "Disabling the NER subtask in which predicted entity labels are used as features lowers performance, showing that the silver-standard entity labels do provide useful features in training. Table 3 : Micro-avg. F 1 -scores for model variations on the SENTiVENT-Event subtasks.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 187,
                        "end": 194,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Results and Discussion",
                "sec_num": "5"
            },
            {
                "text": "Interestingly, there is a large gap in performance on the two datasets while similar model settings were used and while a outwardly similar text genre (news text) was under investigation. We believe there might be several reasons for this performance gap. First of all, the drop in performance can to a certain extent be explained by the DYGIE++ requirement for single-token triggers, which is the unit of investigation in the ACE05 dataset. Many trigger types in our SENTiVENT corpus are differentiated lexically by the combination of a nominal and verbal phrase. In order to obtain single token triggers, a dependency parsing approach was taken in which the syntactic head noun or verb was preferred, introducing ambiguity. A multi-token trigger such as \"conquer consumer spending\" for SALESVOLUME would then lead to a more general and ambiguous trigger \"spending\" which is common in event type EXPENSE). We hypothesize that single-token triggers delineate general-domain ACE05 geo-political event types (LIFE.BE-BORN, LIFE.MARRY, MOVEMENT.TRANSPORT, TRANSACTION.TRANSFER-MONEY) better than our company-specific events.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results and Discussion",
                "sec_num": "5"
            },
            {
                "text": "Finally, due to the higher event-sentence density of our data (cf. Table 1) , there is less global context to learn from, as well as a higher likelihood of overlap in the lexical context (local and global) of triggers and arguments. Table 4 shows the frequency of errors made by the BERT+LSTM system. For event triggers, missing identified triggers (51.7%) constitutes the largest error and 13.6% involves misclassified cases where a trigger was identified but the event type was mistaken (e.g., a REVENUE event is assigned PROFIT/LOSS). Missing event arguments is the largest error type (70.8%) for arguments with misclassification playing a minor role (1.3%). Spurious triggers or arguments account for 34.7% and 27.9%, respectively and they occur when token spans are assigned a label where none is present.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 67,
                        "end": 75,
                        "text": "Table 1)",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 233,
                        "end": 240,
                        "text": "Table 4",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Results and Discussion",
                "sec_num": "5"
            },
            {
                "text": "Missing Spurious Misclassified TRIGGER 51.7% 34.7% 13.6% ARGUMENT 70.8% 27.9% 1.3% 12.4% 9.1% 16.4% 20.1% 34.1% Table 5 : Frequency of trigger errors in manual assessment.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 112,
                        "end": 119,
                        "text": "Table 5",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Error type",
                "sec_num": null
            },
            {
                "text": "We also manually reviewed half of the evaluation test set documents and annotated the errors in more detail (Table 5) . Lexical sparsity errors where triggers are missing or misclassified because they are rare or unseen in training is less of a problem (associated with 9% of errors). Highly specialized contexts or creative language use introduces contextual lexical sparsity. This occurs more frequently with companyspecific news than general news because industry, product, or company-specific terminology is used; e.g., \"Besides its track-tested suspension and race-ready seats, the Edge ST's looks take a dark turn...\" \u2192 true: PRODUCT/SERVICE, pred.: Missing. Ambiguous triggers are also a common source of errors, e.g., \"growth\" is often used for various types related to financial metrics such as PROFIT/LOSS, REVENUE, EXPENSE, etc.: e.g., \"Expect strongest growth from services and Asia.\" \u2192 true: FINANCIALREPORT, pred.: SALESVOLUME. \"Apple made $64.1 billion before taxes in fiscal 2017.\" \u2192 true: PROFIT/LOSS, pred: REVENUE. Another example is \"buy\", which is a trigger often used for MERGER/ACQUISITION but often misclassified as a 'buy/hold/sell' RATING. Better capturing the long context and event coreference should resolve ambiguous trigger mentions as often the preceding context specifies the metric. A large amount of errors (34%) were due to the pre-processing step of converting our ground-truth multi-token triggers into single-tokens. This discards discriminative information regarding type and causes many spurious and missing errors. e.g., \"... organic sales growth projections for this year ...\" where \"sales growth\" is the original annotated SALESVOLUME trigger reduced to \"growth\" introducing class ambiguity which has evidently not been resolved by local or global context. Spurious triggers that plausibly express an event but are absent in the ground-truth are also fairly common (16%). This is not an artifact of low-quality annotations but occurs with non-salient mentions of events which are generic or unspecific in nature and for which a more concrete example is annotated in the direct vicinity. Our conceptualization of economic events includes specificity and relevancy in annotation, which is not wellcaptured by the model, e.g., in \"Below is an analysis of Apple's App Store revenue...\" \"revenue\" is a spurious REVENUE prediction that is not in the ground-truth because it is a generic mention followed by a series of fully-realized events.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 108,
                        "end": 117,
                        "text": "(Table 5)",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Error type",
                "sec_num": null
            },
            {
                "text": "This work's source code, preprocessed replication data in DYGIE-format, and the winning trained model are publicly available at https://osf.io/j63h9 (Jacobs, 2020a) . The original dataset will be freely downloadable at the end of the SENTiVENT project through this repository. Up until then, the fully annotated corpus is available on request for academic research purposes.",
                "cite_spans": [
                    {
                        "start": 149,
                        "end": 164,
                        "text": "(Jacobs, 2020a)",
                        "ref_id": "BIBREF23"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data availability and replication",
                "sec_num": "6"
            },
            {
                "text": "Event extraction is a required step in many data-driven financial tasks in which factual information is needed to capture changes in the real-world. Various general domain fine-grained event extraction corpora are freely available but no economically focused corpus exists. We presented a pilot study on a novel dataset enabling supervised fine-grained extraction of economic events as trigger and argument classification. Using a state-of-the-art information extraction pipeline based on span-based graph propagation of pretrained contextual embeddings, we observed a large drop in performance on our dataset compared to the benchmark ACE05 dataset. After error analysis, we found this is largely caused by missing predictions. For event triggers, many errors are due to ambiguity introduced by the requirement of the DYGIE++ model implementation for single-token triggers. We hypothesize that single-token triggers for the SENTiVENT corpus do not provide sufficient discriminative information for detecting classes which have a large degree of semantic and contextual overlap.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "7"
            },
            {
                "text": "Hence, in future work we will focus on event extraction methods that model arbitrary length triggers. Currently on-going, we are adding \"investor sentiment\" annotations on top of events as well as separate sentiment expression annotations with their targets. These sentiment annotations will allow us to jointly process the \"common-sense\" sentiment of events. We will investigate how extracted event schemata can be used upstream from aspect-based sentiment analysis.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "7"
            },
            {
                "text": "The model weights were obtained from https://github.com/ProsusAI/finBERT.2 The ACE05 scores differ from those reported in Wadden et al. (2019b) as we did not use ensembles. We did use all the same data pre-processing, splits, and single-model settings provided by the authors.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "This work was supported by the Research Foundation Flanders (FWO) under a Ph.D fellowship grant for the SENTiVENT project. We would like to thank anonymous reviewers for their helpful suggestions, as well as David Wadden for publishing and maintaining their model source code and answering questions about the model.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgements",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Finbert: Financial sentiment analysis with pre-trained language models",
                "authors": [
                    {
                        "first": "Dogu",
                        "middle": [],
                        "last": "Araci",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dogu Araci. 2019. Finbert: Financial sentiment analysis with pre-trained language models.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Ontology-Based Information and Event Extraction for Business Intelligence",
                "authors": [
                    {
                        "first": "Ernest",
                        "middle": [],
                        "last": "Arendarenko",
                        "suffix": ""
                    },
                    {
                        "first": "Tuomo",
                        "middle": [],
                        "last": "Kakkonen",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "Artificial Intelligence: Methodology, Systems, and Applications",
                "volume": "7557",
                "issue": "",
                "pages": "89--102",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ernest Arendarenko and Tuomo Kakkonen. 2012. Ontology-Based Information and Event Extraction for Business Intelligence. In Artificial Intelligence: Methodology, Systems, and Applications, volume 7557 of Lecture Notes in Computer Science, pages 89-102. Springer.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Event-based trading: Building superior trading strategies with state-ofthe-art information extraction tools",
                "authors": [
                    {
                        "first": "Zvi",
                        "middle": [],
                        "last": "Ben",
                        "suffix": ""
                    },
                    {
                        "first": "Ami",
                        "middle": [],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Ronen",
                        "middle": [],
                        "last": "Feldman",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zvi Ben Ami and Ronen Feldman. 2017. Event-based trading: Building superior trading strategies with state-of- the-art information extraction tools. SSRN Working Paper 2907600.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Text mining for central banks",
                "authors": [
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Bholat",
                        "suffix": ""
                    },
                    {
                        "first": "Stephen",
                        "middle": [],
                        "last": "Hansen",
                        "suffix": ""
                    },
                    {
                        "first": "Pedro",
                        "middle": [],
                        "last": "Santos",
                        "suffix": ""
                    },
                    {
                        "first": "Cheryl",
                        "middle": [],
                        "last": "Schonhardt-Bailey",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "David Bholat, Stephen Hansen, Pedro Santos, and Cheryl Schonhardt-Bailey. 2015. Text mining for central banks. Available at SSRN 2624811.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Information, trading, and volatility: Evidence from firm-specific news",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Boudoukh",
                        "suffix": ""
                    },
                    {
                        "first": "Ronen",
                        "middle": [],
                        "last": "Feldman",
                        "suffix": ""
                    },
                    {
                        "first": "Shimon",
                        "middle": [],
                        "last": "Kogan",
                        "suffix": ""
                    },
                    {
                        "first": "Matthew P",
                        "middle": [],
                        "last": "Richardson",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jacob Boudoukh, Ronen Feldman, Shimon Kogan, and Matthew P Richardson. 2016. Information, trading, and volatility: Evidence from firm-specific news. SSRN Working Paper 2193667.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Information, trading, and volatility: Evidence from firm-specific news",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Boudoukh",
                        "suffix": ""
                    },
                    {
                        "first": "Ronen",
                        "middle": [],
                        "last": "Feldman",
                        "suffix": ""
                    },
                    {
                        "first": "Shimon",
                        "middle": [],
                        "last": "Kogan",
                        "suffix": ""
                    },
                    {
                        "first": "Matthew",
                        "middle": [],
                        "last": "Richardson",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "The Review of Financial Studies",
                "volume": "32",
                "issue": "3",
                "pages": "992--1033",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jacob Boudoukh, Ronen Feldman, Shimon Kogan, and Matthew Richardson. 2019. Information, trading, and volatility: Evidence from firm-specific news. The Review of Financial Studies, 32(3):992-1033.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Incorporating finegrained events in stock movement prediction",
                "authors": [
                    {
                        "first": "Deli",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Yanyan",
                        "middle": [],
                        "last": "Zou",
                        "suffix": ""
                    },
                    {
                        "first": "Keiko",
                        "middle": [],
                        "last": "Harimoto",
                        "suffix": ""
                    },
                    {
                        "first": "Ruihan",
                        "middle": [],
                        "last": "Bao",
                        "suffix": ""
                    },
                    {
                        "first": "Xuancheng",
                        "middle": [],
                        "last": "Ren",
                        "suffix": ""
                    },
                    {
                        "first": "Xu",
                        "middle": [],
                        "last": "Sun",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the Second Workshop on Economics and Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "31--40",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Deli Chen, Yanyan Zou, Keiko Harimoto, Ruihan Bao, Xuancheng Ren, and Xu Sun. 2019. Incorporating fine- grained events in stock movement prediction. In Proceedings of the Second Workshop on Economics and Natu- ral Language Processing, pages 31-40, Hong Kong, November. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Ace (automatic content extraction) english annotation guidelines for events version 5",
                "authors": [],
                "year": 2005,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Linguistic Data Consortium. 2005. Ace (automatic content extraction) english annotation guidelines for events version 5.4.3.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
                "authors": [
                    {
                        "first": "Jacob",
                        "middle": [],
                        "last": "Devlin",
                        "suffix": ""
                    },
                    {
                        "first": "Ming-Wei",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Kenton",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Kristina",
                        "middle": [],
                        "last": "Toutanova",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "1",
                "issue": "",
                "pages": "4171--4186",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirec- tional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171-4186.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "The automatic content extraction (ace) program-tasks, data, and evaluation",
                "authors": [
                    {
                        "first": "Alexis",
                        "middle": [],
                        "last": "George R Doddington",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Mitchell",
                        "suffix": ""
                    },
                    {
                        "first": "Lance",
                        "middle": [],
                        "last": "Przybocki",
                        "suffix": ""
                    },
                    {
                        "first": "Stephanie",
                        "middle": [],
                        "last": "Ramshaw",
                        "suffix": ""
                    },
                    {
                        "first": "Ralph",
                        "middle": [],
                        "last": "Strassel",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Weischedel",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC'04)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "George R Doddington, Alexis Mitchell, Mark Przybocki, Lance Ramshaw, Stephanie Strassel, and Ralph Weischedel. 2004. The automatic content extraction (ace) program-tasks, data, and evaluation. In Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC'04).",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "PULS: natural language processing for business intelligence",
                "authors": [
                    {
                        "first": "Mian",
                        "middle": [],
                        "last": "Du",
                        "suffix": ""
                    },
                    {
                        "first": "Lidia",
                        "middle": [],
                        "last": "Pivovarova",
                        "suffix": ""
                    },
                    {
                        "first": "Roman",
                        "middle": [],
                        "last": "Yangarber",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the 2016 Workshop on Human Language Technology",
                "volume": "",
                "issue": "",
                "pages": "1--8",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mian Du, Lidia Pivovarova, and Roman Yangarber. 2016. PULS: natural language processing for business intelli- gence. In Proceedings of the 2016 Workshop on Human Language Technology, pages 1-8.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Financial Event Extraction Using Wikipedia-Based Weak Supervision",
                "authors": [
                    {
                        "first": "Ariel",
                        "middle": [],
                        "last": "Liat Ein-Dor",
                        "suffix": ""
                    },
                    {
                        "first": "Orith",
                        "middle": [],
                        "last": "Gera",
                        "suffix": ""
                    },
                    {
                        "first": "Alon",
                        "middle": [],
                        "last": "Toledo-Ronen",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Halfon",
                        "suffix": ""
                    },
                    {
                        "first": "Lena",
                        "middle": [],
                        "last": "Sznajder",
                        "suffix": ""
                    },
                    {
                        "first": "Yonatan",
                        "middle": [],
                        "last": "Dankin",
                        "suffix": ""
                    },
                    {
                        "first": "Yoav",
                        "middle": [],
                        "last": "Bilu",
                        "suffix": ""
                    },
                    {
                        "first": "Noam",
                        "middle": [],
                        "last": "Katz",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Slonim",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the Second Workshop on Economics and Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "10--15",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Liat Ein-Dor, Ariel Gera, Orith Toledo-Ronen, Alon Halfon, Benjamin Sznajder, Lena Dankin, Yonatan Bilu, Yoav Katz, and Noam Slonim. 2019. Financial Event Extraction Using Wikipedia-Based Weak Supervision. In Pro- ceedings of the Second Workshop on Economics and Natural Language Processing, pages 10-15, Stroudsburg, PA, USA. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "The stock sonar sentiment analysis of stocks based on a hybrid approach",
                "authors": [
                    {
                        "first": "Ronen",
                        "middle": [],
                        "last": "Feldman",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Rosenfeld",
                        "suffix": ""
                    },
                    {
                        "first": "Roy",
                        "middle": [],
                        "last": "Bar-Haim",
                        "suffix": ""
                    },
                    {
                        "first": "Moshe",
                        "middle": [],
                        "last": "Fresko",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Twenty-Third IAAI Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ronen Feldman, Benjamin Rosenfeld, Roy Bar-Haim, and Moshe Fresko. 2011a. The stock sonar sentiment analysis of stocks based on a hybrid approach. In Twenty-Third IAAI Conference.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "The Stock Sonar -Sentiment Analysis of Stocks Based on a Hybrid Approach",
                "authors": [
                    {
                        "first": "Ronen",
                        "middle": [],
                        "last": "Feldman",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Rosenfeld",
                        "suffix": ""
                    },
                    {
                        "first": "Roy",
                        "middle": [],
                        "last": "Bar-Haim",
                        "suffix": ""
                    },
                    {
                        "first": "Moshe",
                        "middle": [],
                        "last": "Fresko",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Iaai",
                "volume": "",
                "issue": "",
                "pages": "1642--1647",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ronen Feldman, Benjamin Rosenfeld, Roy Bar-Haim, and Moshe Fresko. 2011b. The Stock Sonar -Sentiment Analysis of Stocks Based on a Hybrid Approach. Iaai, pages 1642-1647.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "AllenNLP: A deep semantic natural language processing platform",
                "authors": [
                    {
                        "first": "Matt",
                        "middle": [],
                        "last": "Gardner",
                        "suffix": ""
                    },
                    {
                        "first": "Joel",
                        "middle": [],
                        "last": "Grus",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Neumann",
                        "suffix": ""
                    },
                    {
                        "first": "Oyvind",
                        "middle": [],
                        "last": "Tafjord",
                        "suffix": ""
                    },
                    {
                        "first": "Pradeep",
                        "middle": [],
                        "last": "Dasigi",
                        "suffix": ""
                    },
                    {
                        "first": "Nelson",
                        "middle": [
                            "F"
                        ],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Matthew",
                        "middle": [],
                        "last": "Peters",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Schmitz",
                        "suffix": ""
                    },
                    {
                        "first": "Luke",
                        "middle": [],
                        "last": "Zettlemoyer",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of Workshop for NLP Open Source Software (NLP-OSS)",
                "volume": "",
                "issue": "",
                "pages": "1--6",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Matt Gardner, Joel Grus, Mark Neumann, Oyvind Tafjord, Pradeep Dasigi, Nelson F. Liu, Matthew Peters, Michael Schmitz, and Luke Zettlemoyer. 2018. AllenNLP: A deep semantic natural language processing platform. In Proceedings of Workshop for NLP Open Source Software (NLP-OSS), pages 1-6, Melbourne, Australia, July. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Event graphs for information retrieval and multi-document summarization",
                "authors": [
                    {
                        "first": "Goran",
                        "middle": [],
                        "last": "Glava\u0161",
                        "suffix": ""
                    },
                    {
                        "first": "Jan\u0161najder",
                        "middle": [],
                        "last": "",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Expert Systems with Applications",
                "volume": "41",
                "issue": "15",
                "pages": "6904--6916",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Goran Glava\u0161 and Jan\u0160najder. 2014. Event graphs for information retrieval and multi-document summarization. Expert Systems with Applications, 41(15):6904-6916.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "An event-extraction approach for business analysis from online Chinese news",
                "authors": [
                    {
                        "first": "Songqiao",
                        "middle": [],
                        "last": "Han",
                        "suffix": ""
                    },
                    {
                        "first": "Xiaoling",
                        "middle": [],
                        "last": "Hao",
                        "suffix": ""
                    },
                    {
                        "first": "Hailiang",
                        "middle": [],
                        "last": "Huang",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Electronic Commerce Research and Applications",
                "volume": "28",
                "issue": "",
                "pages": "244--260",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Songqiao Han, Xiaoling Hao, and Hailiang Huang. 2018. An event-extraction approach for business analysis from online Chinese news. Electronic Commerce Research and Applications, 28:244-260.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Semantics-based information extraction for detecting economic events",
                "authors": [
                    {
                        "first": "Alexander",
                        "middle": [],
                        "last": "Hogenboom",
                        "suffix": ""
                    },
                    {
                        "first": "Frederik",
                        "middle": [],
                        "last": "Hogenboom",
                        "suffix": ""
                    },
                    {
                        "first": "Flavius",
                        "middle": [],
                        "last": "Frasincar",
                        "suffix": ""
                    },
                    {
                        "first": "Kim",
                        "middle": [],
                        "last": "Schouten",
                        "suffix": ""
                    },
                    {
                        "first": "Otto",
                        "middle": [],
                        "last": "Van Der",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Meer",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Multimedia Tools and Applications",
                "volume": "64",
                "issue": "1",
                "pages": "27--52",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alexander Hogenboom, Frederik Hogenboom, Flavius Frasincar, Kim Schouten, and Otto Van Der Meer. 2013. Semantics-based information extraction for detecting economic events. Multimedia Tools and Applications, 64(1):27-52.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "A news event-driven approach for the historical value at risk method",
                "authors": [
                    {
                        "first": "Frederik",
                        "middle": [],
                        "last": "Hogenboom",
                        "suffix": ""
                    },
                    {
                        "first": "Flavius",
                        "middle": [],
                        "last": "Michael De Winter",
                        "suffix": ""
                    },
                    {
                        "first": "Uzay",
                        "middle": [],
                        "last": "Frasincar",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Kaymak",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Expert Systems with Applications",
                "volume": "42",
                "issue": "10",
                "pages": "4667--4675",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Frederik Hogenboom, Michael de Winter, Flavius Frasincar, and Uzay Kaymak. 2015. A news event-driven approach for the historical value at risk method. Expert Systems with Applications, 42(10):4667-4675.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "2017. spacy 2: Natural language understanding with bloom embeddings, convolutional neural networks and incremental parsing",
                "authors": [
                    {
                        "first": "Matthew",
                        "middle": [],
                        "last": "Honnibal",
                        "suffix": ""
                    },
                    {
                        "first": "Ines",
                        "middle": [],
                        "last": "Montani",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Matthew Honnibal and Ines Montani. 2017. spacy 2: Natural language understanding with bloom embeddings, convolutional neural networks and incremental parsing. To appear, 7(1). Dependency parsing model used: en core web lg v2.3.1.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Universal language model fine-tuning for text classification",
                "authors": [
                    {
                        "first": "Jeremy",
                        "middle": [],
                        "last": "Howard",
                        "suffix": ""
                    },
                    {
                        "first": "Sebastian",
                        "middle": [],
                        "last": "Ruder",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics",
                "volume": "1",
                "issue": "",
                "pages": "328--339",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jeremy Howard and Sebastian Ruder. 2018. Universal language model fine-tuning for text classification. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 328-339.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "SENTiVENT: Enabling supervised information extraction of companyspecific events in economic and financial news",
                "authors": [
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Jacobs",
                        "suffix": ""
                    },
                    {
                        "first": "V\u00e9ronique",
                        "middle": [],
                        "last": "Hoste",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gilles Jacobs and V\u00e9ronique Hoste. 2020. SENTiVENT: Enabling supervised information extraction of company- specific events in economic and financial news. Manuscript submitted for publication.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Economic event detection in company-specific news text",
                "authors": [
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Jacobs",
                        "suffix": ""
                    },
                    {
                        "first": "Els",
                        "middle": [],
                        "last": "Lefever",
                        "suffix": ""
                    },
                    {
                        "first": "V\u00e9ronique",
                        "middle": [],
                        "last": "Hoste",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the First Workshop on Economics and Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "1--10",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gilles Jacobs, Els Lefever, and V\u00e9ronique Hoste. 2018. Economic event detection in company-specific news text. In Proceedings of the First Workshop on Economics and Natural Language Processing, pages 1-10.",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Replication data for extracting fine-grained economic events from business news",
                "authors": [
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Jacobs",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gilles Jacobs. 2020a. Replication data for extracting fine-grained economic events from business news, Oct.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "SENTiVENT Event Annotation Guidelines v1",
                "authors": [
                    {
                        "first": "Gilles",
                        "middle": [],
                        "last": "Jacobs",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "",
                "volume": "1",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gilles Jacobs. 2020b. SENTiVENT Event Annotation Guidelines v1.1. Technical report, LT3, Ghent University, jun.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Mining public opinion about economic issues: Twitter and the us presidential election",
                "authors": [
                    {
                        "first": "Amir",
                        "middle": [],
                        "last": "Karami",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "London",
                        "suffix": ""
                    },
                    {
                        "first": "Xiaoyun",
                        "middle": [],
                        "last": "Bennett",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "He",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "International Journal of Strategic Decision Sciences (IJSDS)",
                "volume": "9",
                "issue": "1",
                "pages": "18--28",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Amir Karami, London S Bennett, and Xiaoyun He. 2018. Mining public opinion about economic issues: Twitter and the us presidential election. International Journal of Strategic Decision Sciences (IJSDS), 9(1):18-28.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Subword regularization: Improving neural network translation models with multiple subword candidates",
                "authors": [
                    {
                        "first": "Taku",
                        "middle": [],
                        "last": "Kudo",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "66--75",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Taku Kudo. 2018. Subword regularization: Improving neural network translation models with multiple subword candidates. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Vol- ume 1: Long Papers), pages 66-75, Melbourne, Australia, July. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Ontology-based fuzzy event extraction agent for Chinese e-news summarization",
                "authors": [
                    {
                        "first": "Yea",
                        "middle": [
                            "Juan"
                        ],
                        "last": "Chang Shing Lee",
                        "suffix": ""
                    },
                    {
                        "first": "Zhi W",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Jian",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Expert Systems with Applications",
                "volume": "25",
                "issue": "3",
                "pages": "431--447",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Chang Shing Lee, Yea Juan Chen, and Zhi W Jian. 2003. Ontology-based fuzzy event extraction agent for Chinese e-news summarization. Expert Systems with Applications, 25(3):431-447.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "A classification-based approach to economic event detection in Dutch news text",
                "authors": [
                    {
                        "first": "Els",
                        "middle": [],
                        "last": "Lefever",
                        "suffix": ""
                    },
                    {
                        "first": "V\u00e9ronique",
                        "middle": [],
                        "last": "Hoste",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
                "volume": "",
                "issue": "",
                "pages": "330--335",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Els Lefever and V\u00e9ronique Hoste. 2016. A classification-based approach to economic event detection in Dutch news text. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 330-335, Portoro\u017e, Slovenia, May. European Language Resources Association (ELRA).",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "DEFT Rich ERE Annotation Guidelines: Argument Filler V2.3. Technical report, Linguistic Data Consortium",
                "authors": [],
                "year": 2015,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Linguistic Data Consortium. 2015a. DEFT Rich ERE Annotation Guidelines: Argument Filler V2.3. Technical report, Linguistic Data Consortium, 11.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "DEFT Rich ERE Annotation Guidelines: Events V3.0. Technical report, Linguistic Data Consortium",
                "authors": [],
                "year": 2015,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Linguistic Data Consortium. 2015b. DEFT Rich ERE Annotation Guidelines: Events V3.0. Technical report, Linguistic Data Consortium, 11.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Rich ERE Annotation Guidelines Overview V4.2",
                "authors": [],
                "year": 2016,
                "venue": "Linguistic Data Consortium. Accessed",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Linguistic Data Consortium. 2016. Rich ERE Annotation Guidelines Overview V4.2. Technical report, Linguistic Data Consortium. Accessed: 2018-09-05.",
                "links": null
            },
            "BIBREF32": {
                "ref_id": "b32",
                "title": "Event-based extractive summarization using event semantic relevance from external linguistic resource",
                "authors": [
                    {
                        "first": "Maofu",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Wenjie",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Mingli",
                        "middle": [],
                        "last": "Wu",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings -ALPIT 2007 6th International Conference on Advanced Language Processing and Web Information Technology",
                "volume": "",
                "issue": "",
                "pages": "117--122",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Maofu Liu, Wenjie Li, Mingli Wu, and Jun Hu. 2007. Event-based extractive summarization using event se- mantic relevance from external linguistic resource. Proceedings -ALPIT 2007 6th International Conference on Advanced Language Processing and Web Information Technology, pages 117-122.",
                "links": null
            },
            "BIBREF33": {
                "ref_id": "b33",
                "title": "Accurate information extraction for quantitative financial events",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Hassan",
                        "suffix": ""
                    },
                    {
                        "first": "Vikas",
                        "middle": [
                            "S"
                        ],
                        "last": "Malik",
                        "suffix": ""
                    },
                    {
                        "first": "Huascar",
                        "middle": [],
                        "last": "Bhardwaj",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Fiorletta",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "International Conference on Information and Knowledge Management, Proceedings",
                "volume": "",
                "issue": "",
                "pages": "2497--2500",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hassan H. Malik, Vikas S. Bhardwaj, and Huascar Fiorletta. 2011. Accurate information extraction for quantitative financial events. International Conference on Information and Knowledge Management, Proceedings, pages 2497-2500.",
                "links": null
            },
            "BIBREF34": {
                "ref_id": "b34",
                "title": "Event-based summarization using a centrality-as-relevance model",
                "authors": [
                    {
                        "first": "Lu\u00eds",
                        "middle": [],
                        "last": "Marujo",
                        "suffix": ""
                    },
                    {
                        "first": "Ricardo",
                        "middle": [],
                        "last": "Ribeiro",
                        "suffix": ""
                    },
                    {
                        "first": "Anatole",
                        "middle": [],
                        "last": "Gershman",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Martins De Matos",
                        "suffix": ""
                    },
                    {
                        "first": "Jo\u00e3o",
                        "middle": [
                            "P"
                        ],
                        "last": "Neto",
                        "suffix": ""
                    },
                    {
                        "first": "Jaime",
                        "middle": [],
                        "last": "Carbonell",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Knowledge and Information Systems",
                "volume": "50",
                "issue": "3",
                "pages": "945--968",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lu\u00eds Marujo, Ricardo Ribeiro, Anatole Gershman, David Martins de Matos, Jo\u00e3o P. Neto, and Jaime Carbonell. 2017. Event-based summarization using a centrality-as-relevance model. Knowledge and Information Systems, 50(3):945-968.",
                "links": null
            },
            "BIBREF35": {
                "ref_id": "b35",
                "title": "Walking down wall street with a tablet: A survey of stock market predictions using the web",
                "authors": [
                    {
                        "first": "Michela",
                        "middle": [],
                        "last": "Nardo",
                        "suffix": ""
                    },
                    {
                        "first": "Marco",
                        "middle": [],
                        "last": "Petracco-Giudici",
                        "suffix": ""
                    },
                    {
                        "first": "Min\u00e1s",
                        "middle": [],
                        "last": "Naltsidis",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Journal of Economic Surveys",
                "volume": "30",
                "issue": "2",
                "pages": "356--369",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Michela Nardo, Marco Petracco-Giudici, and Min\u00e1s Naltsidis. 2016. Walking down wall street with a tablet: A survey of stock market predictions using the web. Journal of Economic Surveys, 30(2):356-369.",
                "links": null
            },
            "BIBREF36": {
                "ref_id": "b36",
                "title": "Text mining for market prediction: A systematic review",
                "authors": [
                    {
                        "first": "Saeed",
                        "middle": [],
                        "last": "Arman Khadjeh Nassirtoussi",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Aghabozorgi",
                        "suffix": ""
                    },
                    {
                        "first": "Ying",
                        "middle": [],
                        "last": "Teh",
                        "suffix": ""
                    },
                    {
                        "first": "David Chek Ling",
                        "middle": [],
                        "last": "Wah",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ngo",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Expert Systems with Applications",
                "volume": "41",
                "issue": "16",
                "pages": "7653--7670",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Arman Khadjeh Nassirtoussi, Saeed Aghabozorgi, Teh Ying Wah, and David Chek Ling Ngo. 2014. Text mining for market prediction: A systematic review. Expert Systems with Applications, 41(16):7653-7670.",
                "links": null
            },
            "BIBREF37": {
                "ref_id": "b37",
                "title": "Joint event extraction via recurrent neural networks",
                "authors": [
                    {
                        "first": "Kyunghyun",
                        "middle": [],
                        "last": "Thien Huu Nguyen",
                        "suffix": ""
                    },
                    {
                        "first": "Ralph",
                        "middle": [],
                        "last": "Cho",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Grishman",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "",
                "issue": "",
                "pages": "300--309",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Thien Huu Nguyen, Kyunghyun Cho, and Ralph Grishman. 2016. Joint event extraction via recurrent neural net- works. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Compu- tational Linguistics: Human Language Technologies, pages 300-309, San Diego, California, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF38": {
                "ref_id": "b38",
                "title": "On detecting business event from the headlines and leads of massive online news articles",
                "authors": [
                    {
                        "first": "Yu",
                        "middle": [],
                        "last": "Qian",
                        "suffix": ""
                    },
                    {
                        "first": "Xiongwen",
                        "middle": [],
                        "last": "Deng",
                        "suffix": ""
                    },
                    {
                        "first": "Qiongwei",
                        "middle": [],
                        "last": "Ye",
                        "suffix": ""
                    },
                    {
                        "first": "Baojun",
                        "middle": [],
                        "last": "Ma",
                        "suffix": ""
                    },
                    {
                        "first": "Hua",
                        "middle": [],
                        "last": "Yuan",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Information Processing and Management",
                "volume": "56",
                "issue": "6",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yu Qian, Xiongwen Deng, Qiongwei Ye, Baojun Ma, and Hua Yuan. 2019. On detecting business event from the headlines and leads of massive online news articles. Information Processing and Management, 56(6):102086.",
                "links": null
            },
            "BIBREF39": {
                "ref_id": "b39",
                "title": "Online sentiment analysis in marketing research: a review",
                "authors": [
                    {
                        "first": "Meena",
                        "middle": [],
                        "last": "Rambocas",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Barney",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Pacheco",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Journal of Research in Interactive Marketing",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Meena Rambocas and Barney G Pacheco. 2018. Online sentiment analysis in marketing research: a review. Journal of Research in Interactive Marketing.",
                "links": null
            },
            "BIBREF40": {
                "ref_id": "b40",
                "title": "Bank distress in the news: Describing events through deep learning",
                "authors": [
                    {
                        "first": "Samuel",
                        "middle": [],
                        "last": "R\u00f6nnqvist",
                        "suffix": ""
                    },
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Sarlin",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Neurocomputing",
                "volume": "264",
                "issue": "",
                "pages": "57--70",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Samuel R\u00f6nnqvist and Peter Sarlin. 2017. Bank distress in the news: Describing events through deep learning. Neurocomputing, 264:57-70.",
                "links": null
            },
            "BIBREF41": {
                "ref_id": "b41",
                "title": "Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction",
                "authors": [
                    {
                        "first": "Lei",
                        "middle": [],
                        "last": "Sha",
                        "suffix": ""
                    },
                    {
                        "first": "Feng",
                        "middle": [],
                        "last": "Qian",
                        "suffix": ""
                    },
                    {
                        "first": "Baobao",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Zhifang",
                        "middle": [],
                        "last": "Sui",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "AAAI Conference on Artificial Intelligence",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lei Sha, Feng Qian, Baobao Chang, and Zhifang Sui. 2018. Jointly extracting event triggers and arguments by dependency-bridge rnn and tensor-based argument interaction. In AAAI Conference on Artificial Intelligence.",
                "links": null
            },
            "BIBREF42": {
                "ref_id": "b42",
                "title": "Belgian economic policy uncertainty index: Improvement through text mining",
                "authors": [
                    {
                        "first": "Ellen",
                        "middle": [],
                        "last": "Tobback",
                        "suffix": ""
                    },
                    {
                        "first": "Hans",
                        "middle": [],
                        "last": "Naudts",
                        "suffix": ""
                    },
                    {
                        "first": "Walter",
                        "middle": [],
                        "last": "Daelemans",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "International Journal of Forecasting",
                "volume": "34",
                "issue": "2",
                "pages": "355--365",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ellen Tobback, Hans Naudts, Walter Daelemans, Enric Junqu\u00e9 de Fortuny, and David Martens. 2018. Belgian economic policy uncertainty index: Improvement through text mining. International Journal of Forecasting, 34(2):355 -365.",
                "links": null
            },
            "BIBREF43": {
                "ref_id": "b43",
                "title": "Fine-grained analysis of explicit and implicit sentiment in financial news articles",
                "authors": [
                    {
                        "first": "Marjan",
                        "middle": [],
                        "last": "Van De",
                        "suffix": ""
                    },
                    {
                        "first": "Diane",
                        "middle": [],
                        "last": "Kauter",
                        "suffix": ""
                    },
                    {
                        "first": "V\u00e9ronique",
                        "middle": [],
                        "last": "Breesch",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Hoste",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Expert Systems with Applications",
                "volume": "42",
                "issue": "11",
                "pages": "4999--5010",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Marjan Van De Kauter, Diane Breesch, and V\u00e9ronique Hoste. 2015. Fine-grained analysis of explicit and implicit sentiment in financial news articles. Expert Systems with Applications, 42(11):4999-5010.",
                "links": null
            },
            "BIBREF44": {
                "ref_id": "b44",
                "title": "Dygie++: Span-based system for named entity, relation, and event extraction",
                "authors": [
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Wadden",
                        "suffix": ""
                    },
                    {
                        "first": "Ulme",
                        "middle": [],
                        "last": "Wennberg",
                        "suffix": ""
                    },
                    {
                        "first": "Yi",
                        "middle": [],
                        "last": "Luan",
                        "suffix": ""
                    },
                    {
                        "first": "Hannaneh",
                        "middle": [],
                        "last": "Hajishirzi",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "David Wadden, Ulme Wennberg, Yi Luan, and Hannaneh Hajishirzi. 2019a. Dygie++: Span-based system for named entity, relation, and event extraction. https://github.com/dwadden/dygiepp.",
                "links": null
            },
            "BIBREF45": {
                "ref_id": "b45",
                "title": "Entity, relation, and event extraction with contextualized span representations",
                "authors": [
                    {
                        "first": "David",
                        "middle": [],
                        "last": "Wadden",
                        "suffix": ""
                    },
                    {
                        "first": "Ulme",
                        "middle": [],
                        "last": "Wennberg",
                        "suffix": ""
                    },
                    {
                        "first": "Yi",
                        "middle": [],
                        "last": "Luan",
                        "suffix": ""
                    },
                    {
                        "first": "Hannaneh",
                        "middle": [],
                        "last": "Hajishirzi",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
                "volume": "",
                "issue": "",
                "pages": "5784--5789",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "David Wadden, Ulme Wennberg, Yi Luan, and Hannaneh Hajishirzi. 2019b. Entity, relation, and event extrac- tion with contextualized span representations. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5784-5789, Hong Kong, China, November. Association for Computational Linguis- tics.",
                "links": null
            },
            "BIBREF47": {
                "ref_id": "b47",
                "title": "Bank risk aggregation with forward-looking textual risk disclosures",
                "authors": [
                    {
                        "first": "Lu",
                        "middle": [],
                        "last": "Wei",
                        "suffix": ""
                    },
                    {
                        "first": "Guowen",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Jianping",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Xiaoqian",
                        "middle": [],
                        "last": "Zhu",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "The North American Journal of Economics and Finance",
                "volume": "50",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Lu Wei, Guowen Li, Jianping Li, and Xiaoqian Zhu. 2019. Bank risk aggregation with forward-looking textual risk disclosures. The North American Journal of Economics and Finance, 50:101016.",
                "links": null
            },
            "BIBREF48": {
                "ref_id": "b48",
                "title": "Dcfee: A document-level chinese financial event extraction system based on automatically labeled training data",
                "authors": [
                    {
                        "first": "Hang",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Yubo",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    },
                    {
                        "first": "Kang",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "Yang",
                        "middle": [],
                        "last": "Xiao",
                        "suffix": ""
                    },
                    {
                        "first": "Jun",
                        "middle": [],
                        "last": "Zhao",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of ACL 2018, System Demonstrations",
                "volume": "",
                "issue": "",
                "pages": "50--55",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hang Yang, Yubo Chen, Kang Liu, Yang Xiao, and Jun Zhao. 2018. Dcfee: A document-level chinese financial event extraction system based on automatically labeled training data. In Proceedings of ACL 2018, System Demonstrations, pages 50-55.",
                "links": null
            },
            "BIBREF49": {
                "ref_id": "b49",
                "title": "Stock market prediction via multi-source multiple instance learning",
                "authors": [
                    {
                        "first": "Xi",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "Siyu",
                        "middle": [],
                        "last": "Qu",
                        "suffix": ""
                    },
                    {
                        "first": "Jieyun",
                        "middle": [],
                        "last": "Huang",
                        "suffix": ""
                    },
                    {
                        "first": "Binxing",
                        "middle": [],
                        "last": "Fang",
                        "suffix": ""
                    },
                    {
                        "first": "Philip",
                        "middle": [],
                        "last": "Yu",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "IEEE Access",
                "volume": "6",
                "issue": "",
                "pages": "50720--50728",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Xi Zhang, Siyu Qu, Jieyun Huang, Binxing Fang, and Philip Yu. 2018. Stock market prediction via multi-source multiple instance learning. IEEE Access, 6:50720-50728.",
                "links": null
            },
            "BIBREF50": {
                "ref_id": "b50",
                "title": "Joint entity and event extraction with generative adversarial imitation learning",
                "authors": [
                    {
                        "first": "Tongtao",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "Ji",
                        "middle": [],
                        "last": "Heng",
                        "suffix": ""
                    },
                    {
                        "first": "Avirup",
                        "middle": [],
                        "last": "Sil",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Data Intelligence",
                "volume": "1",
                "issue": "2",
                "pages": "99--120",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tongtao Zhang, Heng Ji, and Avirup Sil. 2019. Joint entity and event extraction with generative adversarial imitation learning. Data Intelligence, 1(2):99-120.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "num": null,
                "text": "Examples of annotated economic event schemata with argument roles. Boldface indicates event trigger spans. Underlining indicates argument spans.",
                "uris": null,
                "type_str": "figure"
            },
            "TABREF0": {
                "num": null,
                "content": "<table><tr><td>675</td><td>4,859</td><td>605</td><td>576</td><td>6040</td></tr></table>",
                "text": "Properties for our SENTiVENT dataset and ACE05 in totals, as well as for the training, development and the test evaluation sets. * indicates that SENTiVENT contains no manually annotated entities.",
                "html": null,
                "type_str": "table"
            },
            "TABREF3": {
                "num": null,
                "content": "<table><tr><td>Trigger Error</td><td>Specialized/Creative</td><td>Lexical</td><td>Plausible</td><td>Ambiguous</td><td>Single-Token</td></tr><tr><td>Type</td><td>Language</td><td>Sparsity</td><td>Spurious</td><td>Trigger</td><td>Pre-processing</td></tr><tr><td>Freq.</td><td/><td/><td/><td/><td/></tr></table>",
                "text": "Frequency of error types of the best system on the test set.",
                "html": null,
                "type_str": "table"
            }
        }
    }
}