{ "paper_id": "M91-1019", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T03:15:38.340915Z" }, "title": "UNL/USL : MUC-3 TEST RESULTS AND ANALYSI S", "authors": [ { "first": "Jitender", "middle": [ "S" ], "last": "Deogun", "suffix": "", "affiliation": {}, "email": "deogun@ferg@vax.unl.edu" }, { "first": "Vijay", "middle": [ "V" ], "last": "Raghavan", "suffix": "", "affiliation": {}, "email": "raghavan@cacs.usl.edu" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "This report describes the results from the required run in addition to the five optional runs for th e Muc-3 task. One purpose of the optional runs is to investigate the precision-recall tradeoff. Two of the optional runs (options 1 and 4) were also submitted to Nose, making up a total of three official runs .", "pdf_parse": { "paper_id": "M91-1019", "_pdf_hash": "", "abstract": [ { "text": "This report describes the results from the required run in addition to the five optional runs for th e Muc-3 task. One purpose of the optional runs is to investigate the precision-recall tradeoff. Two of the optional runs (options 1 and 4) were also submitted to Nose, making up a total of three official runs .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "The experiments are conducted using three different training sets selected from the corpus of 1300 messages for which the key-templates have been manually generated during the first phas e of the Muc-3 project . The Training Set 1 (200 messages) and the Training Set 2 (300 messages ) contain an almost equal number of relevant and non-relevant messages where a message is terme d relevant if it generates at least one template . The Training Set 1 is a proper subset of Training Se t 2 . The Training Set 3 (306 messages) contains only those messages that are relevant and generat e one and only one template . The occurrence distribution of the various incident types for the three training sets are presented in Table 1 . The *ed rows indicate which of the incident types have enough occurrences in the training set to be learnable . Similarly, not all fills associated with other slots ar e learnable .", "cite_spans": [], "ref_spans": [ { "start": 715, "end": 722, "text": "Table 1", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "", "sec_num": null }, { "text": "In any run, only two of the three training sets are used . One of these two training sets is used to develop a rule vector (termed as the optimal_query) that can identify a message as being relevant to the Muc-3 task . The other training set is used to develop concept rule vectors that can identif y which among the various possible slot fills are actually applicable' to a message . Since our syste m mainly deals with slots for which fills come from a predefined set of fills (i.e . these are identified with concepts to be learned), the number of concept rule vectors that pertain to each slot is not to o many .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "The activation value for a slot fill with respect to a test message is computed as the dot product of the concept rule vector and the message representation (as a vector) . For the required and some of the optional runs (options 2, 4, and 5), the system decides that a slot fill applies if its activatio n value with respect to the rule vector forte slot fill is greater than a dynamically generated threshol d T1 . This threshold for a given slot fill is based on the percentage of messages in the training set to which the set fill is applicable and the histogram depicting the distribution of the activation value s 1 \"applicable\" means relevance of fill to a message", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Training Training Type Set 1 Set 2 Set 3 ARSON 10 11 5 ARSON THREAT 0 1 0 MURDER 59 79 13 8 DEATH THREAT 9 12 8 BOMBING 42 63 7 1 BOMB THREAT 0 1 1 KIDNAPPING 15 17 2 5 KIDNAPPING THREAT 0 0 0 HIJACKING 0 0 0 HIJACKING THREAT 0 0 0 ROBBERY 4 8 8 ROBBERY THREAT 0 0 0 ATTACK 23 43 44 ATTEMPTED ARSON 0 0 0 ATTEMPTED MURDER 1 5 7 ATTEMPTED BOMBING 9 12 8 ATTEMPTED KIDNAPPING 0 0 0 ATTEMPTED HIJACKING 0 0 0 ATTEMPTED ROBBERY 0 0 0 Total relevant 124 172 806 Total nonrelevant 78 128 0 incident types for which concept rule vectors are generated by the learning module . in the test set of messages for this slot fill. A second option is to use a zero threshold implying that the slot fill is applicable if the activation value of the corresponding rule vector with respect to th e message representation is positive .", "cite_spans": [], "ref_spans": [ { "start": 9, "end": 559, "text": "Training Type Set 1 Set 2 Set 3 ARSON 10 11 5 ARSON THREAT 0 1 0 MURDER 59 79 13 8 DEATH THREAT 9 12 8 BOMBING 42 63 7 1 BOMB THREAT 0 1 1 KIDNAPPING 15 17 2 5 KIDNAPPING THREAT 0 0 0 HIJACKING 0 0 0 HIJACKING THREAT 0 0 0 ROBBERY 4 8 8 ROBBERY THREAT 0 0 0 ATTACK 23 43 44 ATTEMPTED ARSON 0 0 0 ATTEMPTED MURDER 1 5 7 ATTEMPTED BOMBING 9 12 8 ATTEMPTED KIDNAPPING 0 0 0 ATTEMPTED HIJACKING 0 0 0 ATTEMPTED ROBBERY 0 0 0 Total relevant 124 172 806 Total nonrelevant", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "The training sets and the threshold setting used for official and optional test runs are presente d in Table 2 . The number of templates generated are compared in Table 3 and detailed results in terms of precision, recall, and overgeneration for Option 4 are presented in Table 4 . Since our system placed an emphasis on set list type slot fills, our system's performance, with respect to set fills only, for the various test runs, is summarized in Table 5 . The results for Options 2, 3, and 5 , shown in Tables 3 and 5, are scored at our site rather than by official scorers . Consequently, thes e figures are not completely consistent with those of the other options . In our assessment, the recal l and precision values of our scoring are lower in Table 3 than what they would have been if score d by the official scorers . In contrast, the same options in Table 5 are somewhat inflated compared t o what the official scoring would have yielded . With this disparity in mind, the following observations are made on results from different runs :", "cite_spans": [], "ref_spans": [ { "start": 103, "end": 110, "text": "Table 2", "ref_id": "TABREF1" }, { "start": 163, "end": 170, "text": "Table 3", "ref_id": "TABREF4" }, { "start": 449, "end": 456, "text": "Table 5", "ref_id": "TABREF6" }, { "start": 752, "end": 759, "text": "Table 3", "ref_id": "TABREF4" }, { "start": 861, "end": 868, "text": "Table 5", "ref_id": "TABREF6" } ], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "Required Run (Official-1) The official run generated a large number of templates . This run resulted in a moderate recall and moderate precision .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "Option 1 (Official-2) The run for option 1 does not generate many templates . This option use s a stricter or higher threshold for concepts compared to the Official-1 run . Therefore, thi s method achieves a low recall with a reasonable level of precision . This option sacrifices recal l to improve precision considerably . Required 108 143 62 0 81 46 57 43 5 7 Option 1 104 63 37 0 26 67 36 59 4 1 Option 2 105 125 44 0 81 61 42 35 6 5 Option 3 103 53 26 0 27 77 25 49 5 1 Option 4 107 108 56 0 52 51 52 52 48 Option 5 108 133 54 0 79 54 50 41 two training sets used for optimaLquery versus the other concepts . The effects on recall and precision are insignificant .", "cite_spans": [], "ref_spans": [ { "start": 325, "end": 603, "text": "Required 108 143 62 0 81 46 57 43 5 7 Option 1 104 63 37 0 26 67 36 59 4 1 Option 2 105 125 44 0 81 61 42 35 6 5 Option 3 103 53 26 0 27 77 25 49 5 1 Option 4 107 108 56 0 52 51 52 52 48 Option 5 108 133 54 0 79 54 50 41", "ref_id": "TABREF0" } ], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "Option 3 Option 3 used the same training sets as option 2 but the threshold was set to default . This led to a sharp drop in the total number of templates resulting in a much smaller valu e for recall . But as in option 1, there is significant improvement in precision . Comparing option 2 to option 3 serves the same purpose as comparing option 1 to the required run .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "Option 4 (Official-3) This option used the Training Set 3 to develop the rule vectors for slo t fills . When compared to option 2, this provides an assessment of the effectiveness of replacin g Training Set 1 by Training Set 3 for learning rule vectors for various possible slot fills . This option results in a large number of templates compared to the other options . The use o f Training Set 3 makes the examples in the training set relatively cleaner. The method retains the level of recall roughly at the same level as the Official-1 run and option 2 while improvin g precision .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "Option 5 Option 5 is meant to test the effect of the use of phrases on the development of rule vector .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "The results from this test are similar to those from option 4 . Since we have obtained othe r results (not reported) where the use of phrases is helpful, we feel that the results concerning phrases is not yet conclusive.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training", "sec_num": null }, { "text": "In each experiment, two different training sets are employed . A new (test) message is processe d against the optimaLquery computed from the first training set . The message is processed with respect to the rule vectors corresponding to all the possible slot fills, as computed from the secon d training set . A message is deemed relevant to Muc-database either if it is sufficiently similar t o the optimaLquery or, based on the concepts that are applicable and the rules in the rulebase, th e inference engine evaluates the root concept to be true . Since the second training set is used to develo p rule vectors for different slot fills, it is desirable that the training set messages contain incidence o f all the slot fills . If there are no examples corresponding to a slot fill, the system is unable to develo p a rule vector for that fill and consequently, cannot recognize the occurrence of that slot fill in a new message . The way in which training sets are used enables us to test the effects of not only th e size of the training set, but also the quality in terms of training messages being non-ambiguous an d noise-free .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "EXPLANATION OF TEST SETTING S", "sec_num": null }, { "text": "The second variable in the optional testing -the threshold activation value -is used to selec t or ignore a slot fill. The system compares the representation of each message with respect to th e rule vectors and computes an activation value for the corresponding slot fill . The default activatio n value is taken to be C., that is, a slot fill is deemed applicable if the activation value of its rule vecto r with respect to the message is positive . A precision-recall tradeoff can be achieved by changing th e value of the threshold activation value . If this threshold is lowered, a concept becomes applicable t o more messages resulting in an improvement in recall at the cost of precision .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "EXPLANATION OF TEST SETTING S", "sec_num": null }, { "text": "The team for the Muc-3 project consisted of two professors, three graduate research assistants, an d four part-time programmers . The following graduate students made significant contributions to thi s project : V . K . Elayavalli and Y . Zhang of usL, and S . K . Bhatia of UNL . The bulk of the effort wa s spent on the process for phrase extraction, followed by selection of training set, developing inferenc e engine for the rulebase and the template filler . The learning and use of scoring program also took a considerable amount of time .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "EFFORT", "sec_num": null }, { "text": "The biggest limiting factor was time . Our estimation of the time and manpower for the project wa s also affected, in part, by a lack of participation in MUC-1 and MUC-2 . The project often competed , usually unsuccessfully, for the time of graduate students because of their classes and examinations .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "LIMITING FACTOR S", "sec_num": null }, { "text": "A lot of effort was spent on the extraction and the use of phrases . However, this effort did not provide much contribution as the phrase information was not exploited to its limits . Towards the end, we succeeded in developing interesting techniques for phrase extraction and usage but coul d not realize the benefits due to time constraints . In retrospect, we feel that we should have spent more time on template filler module than on indexing module .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "LIMITING FACTOR S", "sec_num": null }, { "text": "The quality and size of training sets have tremendous effect on the performance of the system . A limitation of hardware affected the size of training set that could be selected . A large training set required larger main memory and computer time for different modules in the project than could b e afforded by the limited computing resources at the two campuses . Again, the computing resource s had to be shared with instructional and other research users which had an adverse effect on th e resources for the project . Limited hardware resources were responsible for our search for a goo d training set . In this project, we were limited to use at most 300 messages in the training set du e to memory and time constraints . Our initial approach was to manually select some messages t o develop the training set . Later, we developed the training set through a program by selecting onl y those messages that addressed exactly one incident type.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "TRAININ G", "sec_num": null }, { "text": "Ideally, the training set should contain enough messages such that all possible set fills are sufficiently represented . If the training set does not contain any message addressing a certain slot fill, the system is incapable of recognizing that slot fill . We also developed a module which could b e used to select a training set by computing the representational similarity of messages in the test set to those in the development set . Unfortunately, the module was not tested well enough to be used for the MuC-3 official testing .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "TRAININ G", "sec_num": null }, { "text": "Nearly all of the system can be used independent of the domain of application . The system automatically learns the rule vectors corresponding to different slot fills and uses these rule vectors t o identify the slot fills in new messages . The only domain dependent part of the system is the rule base that is used to decide the relevance of a message depending on whether certain concepts ar e applicable to the message in a desired combination .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "DOMAIN INDEPENDENCE", "sec_num": null }, { "text": "MuC-3 provided us with a unique opportunity to test our ideas on conceptual classification of documents in the area of message understanding . Our approach is based on the recognition of message contents rather than actually understanding the messages . The recognition of certain patterns in a message allows the system to conclude whether certain subjects are addressed in a message . Th e system is, however, highly sensitive to the selection of a good training set .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CONCLUSIO N", "sec_num": null }, { "text": "In the context of a MUC-like task, the system is capable of recognizing the presence or absenc e of different concepts that correspond to fills drawn from a set of values . Specifically, our system ca n efficiently identify the domain of a message and which among certain salient concepts are addressed . For example, in the case of Option 4 run, the recall and precision associated with the optimal_quer y vector is respectively 0 .78 and 0 .88 (i .e . if the question is whether at least one template shoul d be generated) . The performance, in terms of precision, is also impressive in certain slots such a s INCIDENT TYPE(S) and effects on PHYSICAL OR HUMAN TARGETS . Furthermore, the concept rul e vectors are found to be successful in identifying relevant paragraphs within messages . In many application environments, such capabilities may be adequate. Furthermore, our system may be use d as a front-end to a comprehensive message understanding system .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CONCLUSIO N", "sec_num": null }, { "text": "The system has only a limited ability to identify fills that are of string type appearing in th e messages . Phrase extraction, combined with the locality of information, was particularly useful i n filling certain slots that require string fills. The process to extract and use phrase information can be exploited to a greater extent than has been done in the present system .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CONCLUSIO N", "sec_num": null }, { "text": "The system can be improved by the following enhancements. From a domain independent viewpoint, the system can benefit from a more robust procedure for training set selection . Moreover, th e process to extract and use phrase information can be exploited to a greater extent . For improvin g the performance in the current domain, the template filler module can be modified by taking int o account the information regarding dependencies between different slot fills . In general, much more effort is needed in designing the template filler module .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CONCLUSIO N", "sec_num": null }, { "text": "The papers in this section, which were prepared by each of the fifteen site s that completed the MUC-3 evaluation, describe the systems that were tested . Th e papers are intended not only to outline each system's architecture but also t o provide the reader with an understanding of the effectiveness of the technique s that were used to handle the particular phenomena found in the MUC-3 corpus . T o make the discussion of these techniques concrete, most of the sites make specifi c reference to some of the phenomena found in message TST1-MUC3-0099 from th e dry-run test set and discuss their system's handling of those phenomena . The ful l text and answer key templates for that message are found in appendix H of th e proceedings .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "PART III : SYSTEM DESCRIPTION S", "sec_num": null }, { "text": "The sites were asked to include the following pieces of information in this paper : things system got righ t things system got wrong", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "PART III : SYSTEM DESCRIPTION S", "sec_num": null } ], "back_matter": [], "bib_entries": {}, "ref_entries": { "FIGREF0": { "num": null, "type_str": "figure", "text": "Background : how/for what the system was developed, an d how much time was spent on the system before MUC-3 * Explanation of the modules of the syste m * Explanation of flow of control (interleaved/sequential/ . . . ) * Explanation (without system-specific jargon) of processing stages : Identification of relevant texts and paragraph s Lexical look-up (example of output and lexicon ) -Syntactic analysis (example of output and grammar ) Semantic analysis (example of output and semantic rules ) in template, with an explanation of interesting things :", "uris": null }, "TABREF0": { "type_str": "table", "text": "Frequency of incident types in", "html": null, "num": null, "content": "
Require dTraining Set 1Training Set 2T i
Option 1Training Set 1Training Set 20
Option 2Training Set 2Training Set 1T i
Option 3Training Set 2Training Set 10
Option 4Training Set 2Training Set 3T i
Option 5Training Set 2Training Set 3TiNo phrase s
" }, "TABREF1": { "type_str": "table", "text": "Different parameter setting s", "html": null, "num": null, "content": "" }, "TABREF4": { "type_str": "table", "text": "Results from different tests for Template-id slo t", "html": null, "num": null, "content": "
SlotRecallPrecisionOvergeneratio n
Template-Id525248
Incident-Type50950
Category385820
Org-Perps251876
Perp-Confidence41054
Phys-Target-Types1500
Human-Target-Types7434
Incident-Location252655
Phys-Effects11331 7
Human-Effects42828
Matched only274145
Matched/Missing154145
All Templates152389
Set Fills Only185720
" }, "TABREF5": { "type_str": "table", "text": "", "html": null, "num": null, "content": "
: Detailed results for Option 4
Summary a et ills
TestPossibleActualCorrectIncorrectSpuriousMissingRecallPrecisionOvergeneration
Required57022367387342016403 3
Option 15476135394958891 5
Option 255717210754042520642 5
Option 35444837085047771 7
Option 456817980203542418572 0
Option 5571208139441404256820
" }, "TABREF6": { "type_str": "table", "text": "Results from different tests based on SET FILLS ONLY row", "html": null, "num": null, "content": "" } } } }