DLJUST at SemEval-2021 Task 7: Hahackathon: Linking Humor and Offense

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1114–1119 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1114 DLJUST at SemEval-2021 Task 7: Hahackathon: Linking Humor and Offense Hani Al-Omari Jordan University of Science and Technology Computer Information Systems Department Irbid, Jordan alomarihani1997@gmail.com Isra’a AbedulNabi Jordan University of Science and Technology Computer Information Systems Department Irbid, Jordan israha95@gmail.com Rehab Duwairi Jordan University of Science and Technology Computer Information Systems Department Irbid, Jordan rehab@just.edu.jo Abstract Humor detection and rating poses interest- ing linguistic challenges to NLP; it is highly subjective depending on the perceptions of a joke and the context in which it is used. This paper utilizes and compares transform- ers models; BERT base and Large, BERTweet, RoBERTa base and Large, and RoBERTa base irony, for detecting and rating humor and of- fense. The proposed models, where given a text in cased and uncased type obtained from SemEval-2021 Task7: HaHackathon: Link- ing Humor and Offense Across Different Age Groups. The highest scored model for the first subtask: Humor Detection, is BERTweet base cased model with 0.9540 F1-score, for the second subtask: Average Humor Rating Score, it is BERT Large cased with the min- imum RMSE of 0.5555, for the fourth sub- task: Average Offensiveness Rating Score, it is BERTweet base cased model with minimum RMSE of 0.4822. 1 Introduction SemEval 2021 Task7 is constructed to detect and rate the humor and offense inside jokes in the En- glish language (Meaney et al., 2021). Humor is an essential aspect of strengthening human communi- cation and relations. However, the interpretation of humor differs based on the perceptions of a joke and the context in which it is used. In 2012, the Hu- man Rights Commission found the most commonly reported form of harassment in Australia was sexist or offensive jokes(the, 2012), humor appreciation; is a highly subjective phenomenon as a sense of humor varies from person to person depending on factors such as age, gender, and socio-economic status. In this task, data labels and ratings were collected from a balanced set of age groups from 18-70. Moreover, annotators represent a variety of genders, political stances, and income levels. The automatic detection of linguistic elements in natural language texts such as aggression, hu- mor, irony, and sarcasm has drawn attention to research communities (Davidov et al., 2010). Several studies and experiments have been per- formed to develop and improve humor detection systems(Annamoradnejad, 2021)(Winters and De- lobelle, 2020)(Sane et al., 2019)(Mao and Liu, 2019)(Chen and Soo, 2018). BERT language model(Annamoradnejad, 2021)(Devlin et al., 2019) showcase the highest results compared to all other works. This paper aims to utilize the Transformers models; BERT base and Large (Devlin et al., 2019), BERTweet (Nguyen et al., 2020), RoBERTa base and Large (Liu et al., 2019), and RoBERTa base irony (Barbieri et al., 2020) for humor detection, humor rating, and offense rating using the dataset obtained from SemEval-2021 Task7 that contains training, development, and test data. Our contri- butions are: Prepossessing text …

Content truncated to 500 words