# coding=utf-8 # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """The TID-8 (The Inherent-Disagreement-8 datasets) benchmark""" import json import os import datasets _TID_8_CITATION = """\ @inproceedings{deng2023tid8, title={You Are What You Annotate: Towards Better Models through Annotator Representations}, author={Deng, Naihao and Liu, Siyang and Zhang, Frederick Xinliang and Wu, Winston and Wang, Lu and Mihalcea, Rada}, booktitle={Findings of EMNLP 2023}, year={2023} } Note that each TID-8 dataset has its own citation. Please see the source to get the correct citation for each contained dataset. """ _TID_8_DESCRIPTION = """\ TID-8 is a new benchmark focused on the task of letting models learn from data that has inherent disagreement. """ _FIA_DESCRIPTION = """\ Friends QIA (Damgaard et al., 2021) is a corpus of classifying indirect answers to polar questions.""" _PEJ_DESCRIPTION = """\ Pejorative (Dinu et al., 2021) classifies whether Tweets contain words that are used pejora- tively. By definition, pejorative words are words or phrases that have negative connotations or that are intended to disparage or belittle.""" _HSB_DESCRIPTION = """\ HS-Brexit (Akhtar et al., 2021) is an abu- sive language detection corpus on Brexit belonging to two distinct groups: a target group of three Mus- lim immigrants in the UK, and a control group of three other individuals.""" _MDA_DESCRIPTION = """\ MultiDomain Agreement (Leonardelli et al., 2021) is a hate speech classification dataset of English tweets from three domains of Black Lives Matter, Election, and Covid-19, with a particular focus on tweets that potentially leads to disagree- ment.""" _GOE_DESCRIPTION = """\ Go Emotions (Demszky et al., 2020) is a fine-grained emotion classification corpus of care- fully curated comments extracted from Reddit. We group emotions into four categories following sen- timent level divides in the original paper.""" _HUM_DESCRIPTION = """\ Humor (Simpson et al., 2019) is a corpus of online texts for pairwise humorousness compari- son""" _COM_DESCRIPTION = """\ CommitmentBank (De Marneffe et al., 2019) is an NLI dataset. It contains naturally oc- curring discourses whose final sentence contains a clause-embedding predicate under an entailment canceling operator (question, modal, negation, an- tecedent of conditional).""" _SNT_DESCRIPTION = """\ Sentiment Analysis (Díaz et al., 2018) is a sentiment classification dataset originally used to detect age-related sentiments.""" _ANNOTATION_SPLIT_DESCRIPTION = """\ Annotation Split: We split the annotations for each annotator into train and test set. In other words, the same set of annotators appear in both train, (val), and test sets. For datasets that have splits originally, we follow the original split and remove datapoints in test sets that are annotated by an annotator who is not in the training set. For datasets that do not have splits originally, we split the data into train and test set for convenience, you may further split the train set into a train and val set. """ _ANNOTATOR_SPLIT_DESCRIPTION = """\ Annotator Split: We split annotators into train and test set. In other words, a different set of annotators would appear in train and test sets. We split the data into train and test set for convenience, you may consider further splitting the train set into a train and val set for performance validation. """ _FIA_CITATION = """\ @inproceedings{damgaard-etal-2021-ill, title = "{``}{I}{'}ll be there for you{''}: The One with Understanding Indirect Answers", author = "Damgaard, Cathrine and Toborek, Paulina and Eriksen, Trine and Plank, Barbara", booktitle = "Proceedings of the 2nd Workshop on Computational Approaches to Discourse", month = nov, year = "2021", address = "Punta Cana, Dominican Republic and Online", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2021.codi-main.1", doi = "10.18653/v1/2021.codi-main.1", pages = "1--11", }""" _PEJ_CITATION = """\ @inproceedings{dinu-etal-2021-computational-exploration, title = "A Computational Exploration of Pejorative Language in Social Media", author = "Dinu, Liviu P. and Iordache, Ioan-Bogdan and Uban, Ana Sabina and Zampieri, Marcos", booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021", month = nov, year = "2021", address = "Punta Cana, Dominican Republic", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2021.findings-emnlp.296", doi = "10.18653/v1/2021.findings-emnlp.296", pages = "3493--3498" }""" _HSB_CITATION = """\ @article{akhtar2021whose, title={Whose opinions matter? perspective-aware models to identify opinions of hate speech victims in abusive language detection}, author={Akhtar, Sohail and Basile, Valerio and Patti, Viviana}, journal={arXiv preprint arXiv:2106.15896}, year={2021} }""" _MDA_CITATION = """\ @inproceedings{leonardelli-etal-2021-agreeing, title = "Agreeing to Disagree: Annotating Offensive Language Datasets with Annotators{'} Disagreement", author = "Leonardelli, Elisa and. Menini, Stefano and Palmero Aprosio, Alessio and Guerini, Marco and Tonelli, Sara", booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing", month = nov, year = "2021", address = "Online and Punta Cana, Dominican Republic", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2021.emnlp-main.822", pages = "10528--10539", }""" _GOE_CITATION = """\ @inproceedings{demszky-etal-2020-goemotions, title = "{G}o{E}motions: A Dataset of Fine-Grained Emotions", author = "Demszky, Dorottya and Movshovitz-Attias, Dana and Ko, Jeongwoo and Cowen, Alan and Nemade, Gaurav and Ravi, Sujith", booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics", month = jul, year = "2020", address = "Online", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2020.acl-main.372", doi = "10.18653/v1/2020.acl-main.372", pages = "4040--4054" }""" _HUM_CITATION = """\ @inproceedings{simpson-etal-2019-predicting, title = "Predicting Humorousness and Metaphor Novelty with {G}aussian Process Preference Learning", author = "Simpson, Edwin and Do Dinh, Erik-L{\^a}n and Miller, Tristan and Gurevych, Iryna", booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics", month = jul, year = "2019", address = "Florence, Italy", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/P19-1572", doi = "10.18653/v1/P19-1572", pages = "5716--5728" }""" _COM_CITATION = """\ @inproceedings{de2019commitmentbank, title={The commitmentbank: Investigating projection in naturally occurring discourse}, author={De Marneffe, Marie-Catherine and Simons, Mandy and Tonhauser, Judith}, booktitle={proceedings of Sinn und Bedeutung}, volume={23}, number={2}, pages={107--124}, year={2019} }""" _SNT_CITATION = """\ @inproceedings{diaz2018addressing, title={Addressing age-related bias in sentiment analysis}, author={D{\'\i}az, Mark and Johnson, Isaac and Lazar, Amanda and Piper, Anne Marie and Gergle, Darren}, booktitle={Proceedings of the 2018 chi conference on human factors in computing systems}, pages={1--14}, year={2018} }""" class TID8Config(datasets.BuilderConfig): """BuilderConfig for TID-8.""" def __init__(self, features, data_url, citation, url, label_classes=("False", "True"),\ task=None, **kwargs): """BuilderConfig for TID-8. Args: features: `list[string]`, list of the features that will appear in the feature dict. Should not include "label". data_url: `string`, url to download the zip file from. citation: `string`, citation for the data set. url: `string`, url for information about the data set. label_classes: `list[string]`, the list of classes for the label if the label is present as a string. Non-string labels will be cast to either 'False' or 'True'. **kwargs: keyword arguments forwarded to super. """ # Version history: # 0.0.2: Initial version. super(TID8Config, self).__init__(version=datasets.Version("1.0.3"), **kwargs) self.features = features self.label_classes = label_classes self.data_url = data_url self.citation = citation self.url = url self.task = task BASE_URL = "https://raw.githubusercontent.com/MichiganNLP/tid8-dataset/main/huggingface-data" class TID8Glue(datasets.GeneratorBasedBuilder): """The TID-8 benchmark.""" BUILDER_CONFIGS = [ TID8Config( name="friends_qia-ann", description=_FIA_DESCRIPTION, features=["Season", "Episode", "Category", "Q_person", \ "A_person", "Q_original", "Q_modified", "A_modified", "Annotation_1", "Annotation_2", \ "Annotation_3", "Goldstandard"], label_classes=["1", "2", "3", "4", "5"], data_url=f"{BASE_URL}/friends_qia-ann.zip", citation=_FIA_CITATION, url="https://github.com/friendsQIA/Friends_QIA", task="indirect_ans" ), TID8Config( name="pejorative-ann", description=_PEJ_DESCRIPTION, features=["pejor_word", "word_definition", "annotator-1", "annotator-2", "annotator-3"], label_classes=["pejorative", "non-pejorative", "undecided"], data_url=f"{BASE_URL}/pejorative-ann.zip", citation=_PEJ_CITATION, url="https://nlp.unibuc.ro/resources.html", task="pejorative" ), TID8Config( name="hs_brexit-ann", description=_HSB_DESCRIPTION, features=["other annotations"], # List label_classes=["hate_speech", "not_hate_speech"], data_url=f"{BASE_URL}/hs_brexit-ann.zip", citation=_HSB_CITATION, url="https://le-wi-di.github.io/", task="hs_brexit" ), TID8Config( name="md-agreement-ann", description=_MDA_DESCRIPTION, features=["task", "original_id", "domain"], label_classes=["offensive_speech", "not_offensive_speech"], data_url=f"{BASE_URL}/md-agreement-ann.zip", citation=_MDA_CITATION, url="https://le-wi-di.github.io/", task="offensive" ), TID8Config( name="goemotions-ann", description=_GOE_DESCRIPTION, features=["author", "subreddit", "link_id", "parent_id", "created_utc", "rater_id", \ "example_very_unclear", "admiration", "amusement", "anger", "annoyance", "approval", \ "caring", "confusion", "curiosity", "desire", "disappointment", "disapproval", \ "disgust", "embarrassment", "excitement", "fear", "gratitude", "grief", "joy", \ "love", "nervousness", "optimism", "pride", "realization", "relief", "remorse", \ "sadness", "surprise", "neutral"], label_classes=["positive", "ambiguous", "negative", "neutral"], data_url=f"{BASE_URL}/goemotions-ann.zip", citation=_GOE_CITATION, url="https://github.com/google-research/google-research/tree/master/goemotions", task="emotion" ), TID8Config( name="humor-ann", description=_HUM_DESCRIPTION, features=["text_a", "text_b"], label_classes=["B", "X", "A"], data_url=f"{BASE_URL}/humor-ann.zip", citation=_HUM_CITATION, url="https://github.com/ukplab/acl2019-GPPL-humour-metaphor", task="humor" ), TID8Config( name="commitmentbank-ann", description=_COM_DESCRIPTION, ## weak_labels are a list features=["HitID", "Verb", "Context", "Prompt", "Target", "ModalType", \ "Embedding", "MatTense", "weak_labels"], label_classes=["0", "1", "2", "3", "-3", "-1", "-2"], data_url=f"{BASE_URL}/commitmentbank-ann.zip", citation=_COM_CITATION, url="https://github.com/mcdm/CommitmentBank", task="certainty" ), TID8Config( name="sentiment-ann", description=_SNT_DESCRIPTION, features=[], label_classes=["Neutral", "Somewhat positive", "Very negative", "Somewhat negative", "Very positive"], data_url=f"{BASE_URL}/sentiment-ann.zip", citation=_SNT_CITATION, url="https://dataverse.harvard.edu/dataverse/algorithm-age-bias", task="sentiment" ), TID8Config( name="friends_qia-atr", description=_FIA_DESCRIPTION, features=["Season", "Episode", "Category", "Q_person", \ "A_person", "Q_original", "Q_modified", "A_modified", "Annotation_1", "Annotation_2", \ "Annotation_3", "Goldstandard"], label_classes=["1", "2", "3", "4", "5"], data_url=f"{BASE_URL}/friends_qia-atr.zip", citation=_FIA_CITATION, url="https://github.com/friendsQIA/Friends_QIA", task="indirect_ans" ), TID8Config( name="pejorative-atr", description=_PEJ_DESCRIPTION, features=["pejor_word", "word_definition", "annotator-1", "annotator-2", "annotator-3"], label_classes=["pejorative", "non-pejorative", "undecided"], data_url=f"{BASE_URL}/pejorative-atr.zip", citation=_PEJ_CITATION, url="https://nlp.unibuc.ro/resources.html", task="pejorative" ), TID8Config( name="hs_brexit-atr", description=_HSB_DESCRIPTION, features=["other annotations"], # List label_classes=["hate_speech", "not_hate_speech"], data_url=f"{BASE_URL}/hs_brexit-atr.zip", citation=_HSB_CITATION, url="https://le-wi-di.github.io/", task="hs_brexit" ), TID8Config( name="md-agreement-atr", description=_MDA_DESCRIPTION, features=["task", "original_id", "domain"], label_classes=["offensive_speech", "not_offensive_speech"], data_url=f"{BASE_URL}/md-agreement-atr.zip", citation=_MDA_CITATION, url="https://le-wi-di.github.io/", task="offensive" ), TID8Config( name="goemotions-atr", description=_GOE_DESCRIPTION, features=["author", "subreddit", "link_id", "parent_id", "created_utc", "rater_id", \ "example_very_unclear", "admiration", "amusement", "anger", "annoyance", "approval", \ "caring", "confusion", "curiosity", "desire", "disappointment", "disapproval", \ "disgust", "embarrassment", "excitement", "fear", "gratitude", "grief", "joy", \ "love", "nervousness", "optimism", "pride", "realization", "relief", "remorse", \ "sadness", "surprise", "neutral"], label_classes=["positive", "ambiguous", "negative", "neutral"], data_url=f"{BASE_URL}/goemotions-atr.zip", citation=_GOE_CITATION, url="https://github.com/google-research/google-research/tree/master/goemotions", task="emotion" ), TID8Config( name="humor-atr", description=_HUM_DESCRIPTION, features=["text_a", "text_b"], label_classes=["B", "X", "A"], data_url=f"{BASE_URL}/humor-atr.zip", citation=_HUM_CITATION, url="https://github.com/ukplab/acl2019-GPPL-humour-metaphor", task="humor" ), TID8Config( name="commitmentbank-atr", description=_COM_DESCRIPTION, # weak_labels are a list features=["HitID", "Verb", "Context", "Prompt", "Target", "ModalType", \ "Embedding", "MatTense", "weak_labels"], label_classes=["0", "1", "2", "3", "-3", "-1", "-2"], data_url=f"{BASE_URL}/commitmentbank-atr.zip", citation=_COM_CITATION, url="https://github.com/mcdm/CommitmentBank", task="certainty" ), TID8Config( name="sentiment-atr", description=_SNT_DESCRIPTION, features=[], label_classes=["Neutral", "Somewhat positive", "Very negative", "Somewhat negative", "Very positive"], data_url=f"{BASE_URL}/sentiment-atr.zip", citation=_SNT_CITATION, url="https://dataverse.harvard.edu/dataverse/algorithm-age-bias", task="sentiment" ), ] def _info(self): features = {} for feature in self.config.features: if "commitmentbank" in self.config.name and feature == "weak_labels": features[feature] = datasets.features.Sequence(datasets.Value("string")) elif "hate_speech_brexit" in self.config.name and feature == "other annotations": features[feature] = datasets.features.Sequence(datasets.Value("string")) else: features[feature] = datasets.Value("string") features["question"] = datasets.Value("string") features["uid"] = datasets.Value("string") features["id"] = datasets.Value("int32") features["annotator_id"] = datasets.Value("string") features["answer"] = datasets.Value("string") features["answer_label"] = datasets.features.ClassLabel(names=self.config.label_classes) additional_split_descr = None if self.config.name.endswith("-ann"): additional_split_descr = _ANNOTATION_SPLIT_DESCRIPTION else: assert self.config.name.endswith("-atr") additional_split_descr = _ANNOTATOR_SPLIT_DESCRIPTION return datasets.DatasetInfo( description=_TID_8_DESCRIPTION + "\n" + self.config.description + "\n" + additional_split_descr, features=datasets.Features(features), homepage=self.config.url, citation=self.config.citation + "\n" + _TID_8_CITATION, ) def _split_generators(self, dl_manager): dl_dir = dl_manager.download_and_extract(self.config.data_url) or "" splits = [] if self.config.name in {"friends_qia-ann", "multi-domain-agreement-ann"}: splits.append( datasets.SplitGenerator( name=datasets.Split.VALIDATION, gen_kwargs={ "data_file": os.path.join(dl_dir, self.config.name, "dev.jsonl"), "split": datasets.Split.VALIDATION, }, ), ) splits.extend([ datasets.SplitGenerator( name=datasets.Split.TRAIN, gen_kwargs={ "data_file": os.path.join(dl_dir, self.config.name, "train.jsonl"), "split": datasets.Split.TRAIN, }, ), datasets.SplitGenerator( name=datasets.Split.TEST, gen_kwargs={ "data_file": os.path.join(dl_dir, self.config.name, "test.jsonl"), "split": datasets.Split.TEST, }, ), ]) return splits def _generate_examples(self, data_file, split): with open(data_file, encoding="utf-8") as f: for i, line in enumerate(f): row = json.loads(line) example = { "id": row["id"], "uid": row["uid"], "answer": row[self.config.task], "answer_label": row[self.config.task], "annotator_id": row["respondent_id"], "question": row["sentence"] } for feature in self.config.features: try: example[feature] = row[feature] except Exception: print(row) yield i, example