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

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1108–1113 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1108 Tsia at SemEval-2021 Task 7: Detecting and Rating Humor and Offense Zhengyi Guan, Xiaobing Zhou* School of Information Science and Engineering Yunnan University, Yunnan, P.R. China *Corresponding author:zhouxb@ynu.edu.com Abstract This paper describes our contribution to SemEval-2021 Task 7:Detecting and Rat- ing Humor and Offense.This task contains two sub-tasks, sub-task 1 and sub-task 2. Among them, sub-task 1 contains three sub-tasks, sub-task 1a ,sub-task 1b and sub-task 1c.Sub-task 1a is to predict if the text would be considered humorous.Sub- task 1c is described as follows: if the text is classed as humorous, predict if the hu- mor rating would be considered controver- sial, i.e. the variance of the rating between annotators is higher than the median.we combined three pre-trained model with CNN to complete these two classification sub-tasks.Sub-task 1b is to judge the de- gree of humor.Sub-task 2 aims to predict how offensive a text would be with values between 0 and 5.We use the idea of regres- sion to deal with these two sub-tasks.We analyze the performance of our method and demonstrate the contribution of each component of our architecture.We have achieved good results under the combina- tion of multiple pre-training models and optimization methods. 1 Introduction Humor is an intellectual activity that can cause cer- tain emotions in human thinking. Humor is not only very important but also very common in daily life. People’s research on humor has involved many fields such as psychology, sociology, linguistics and so on.Of course, it also has special value for the research of computing languages.Because of its complexity and inherent subjectivity, the de- velopment of automatic humor recognition and assessment poses a great challenge in Computa- tional Linguistics, and therefore is a popular sub- ject in various shared task competitions.(Dick et al., 2020)However, we must also recognize the difficul- ties of humor research.First of all, although humans can easily judge whether a sentence is humorous in daily life, but due to humor is restricted by geog- raphy, environment, social background and other aspects, we usually not only pay attention to this sentence or whether the matter is humorous, We have to figure out how humorous or funny this con- text is? In other words, we pay more attention to its degree of humor which is not so easy for computers.And because humor is affected by the environment, different people have different under- standing of humor. Just like sometimes your humor is based on the suffering of others. Things you find funny, but others don’t necessarily find them funny. In other words, humor is controversial. So we have to determine the specific humor rate. This task is to take a median as the criterion for humor. It is also difficult to judge the humor of this dispute by computer language. More recently, some humorous sentences can also have derogatory and offensive elements. Whether humor can cause offense is also one …

Content truncated to 500 words