Comparison and Analysis of Humorous Factors in Text

Humor, and Irony are two parts that have challenged machine learning for many years. In this paper, to improve the machine performance in the semantic field, we analyze the data in SemEval 2020 and consider SemEval 2021 experiments as a comparison. We aim to find out the factor that could affect the humor grade level. For example, some statistic methods, regression analysis, were used in the paper to test some of the hypotheses to improve many applications in AI fields, for instance, human-machine interaction. We show how some factors affect the humor levels, such as the relationship between original words and substitute words, the word number of the headlines, some elements in the headlines. Finally, some basic conclusion about some factors that could affect humor level was concluded. Indeed, the relevance of two words is a factor that could affect the humor level, while the word number of the headlines could not be a factor in this field. In this way, cosine distance may be a shortcut for improving machines' ability to analyze the meaning of sentences.