Grenzlinie at SemEval-2021 Task 7: Detecting and Rating Humor and Offense

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 281–285 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 281 Grenzlinie at SemEval-2021 Task 7: Detecting and Rating Humor and Offense Renyuan Liu Yunnan University, Yunnan, P.R. China bluewind159@qq.com Xiaobing Zhou * Yunnan University, Yunnan, P.R. China zhouxb@ynu.edu.com Abstract This paper introduces the result of Team Gren- zlinie’s experiment in SemEval-2021 task 7: HaHackathon: Detecting and Rating Humor and Offense in English. This task has two sub- tasks. Subtask1 includes the humor detection task, the humor rating prediction task, and the humor controversy detection task. Subtask2 is an offensive rating prediction task. Detec- tion task is a binary classification task, and the rating prediction task is a regression task be- tween 0 to 5. 0 means the task is not humor- ous or not offensive, 5 means the task is very humorous or very offensive. For all the tasks, this paper chooses RoBERTa as the pre-trained model. In classification tasks, Bi-LSTM and adversarial training are adopted. In the regres- sion task, the Bi-LSTM is also adopted. And then we propose a new approach named com- pare method. Finally, our system achieves an F1-score of 95.05% in the humor detection task, F1-score of 61.74% in the humor contro- versy detection task, 0.6143 RMSE in humor rating task, 0.4761 RMSE in the offensive rat- ing task on the test datasets. 1 Introduction Humorous is one kind most interesting, most has the power, most has the universal significance trans- mission art. Therefore, humor is one of the ways to improve the quality of daily conversation. In the field of natural language processing, how to make the computer learn humor and improve the qual- ity of human-computer interaction is an important problem. The previous researches task was only to input the humorous corpus into the deep learning network and let the algorithm learn how to generate humorous dialogue. In this case, the sentences are often problematic. Because humor is an abstract concept, in different situations, the degree of humor and the way of humor will be different. Therefore, before the computer learns to generate humorous sentences, it is an important task for the computer to understand humor and distinguish different de- grees and forms of humor. This paper mainly discusses how to identify these humorous sentences automatically. In SemEval-2021 task 7, subtask1 includes the hu- mor detection task, the humor rating predicts task and the humor controversy detection task (Meaney et al., 2021). Subtask2 is an offensive rating predict task. In the detection task, the Bi-LSTM and adver- sarial training (Tram`er et al., 2017) is adopted, we also try to use FocalLoss to solve the data unbal- ance problem. In the regression task, the Bi-LSTM is also adopted. And then we propose a new method named compare method is also adopted. The rest of the paper is as follows: Section 2 briefly introduces the related work. Section 3 de- scribes the optimization approach to be used in detail. Section 4 …

Content truncated to 500 words