Study on the Classification of Chinese Black Humor Texts based on Transformer, Word2Vec, and BERT Techniques

With the rapid development of information technology, the number of Chinese black humor texts has surged. Efficiently and accurately classifying their sentiment orientation has become a key research focus in natural language processing (NLP). This study delves into the performance of three models-BERT, Transformer, and Word2Vec-on the task of Chinese black humor text classification. By collecting and organizing relevant corpora from online forums, black humor expressions, and GitHub repositories, two types of datasets were constructed: single-sentence and question-answer datasets. Comparative analyses of the three models reveal that the BERT model, leveraging its bidirectional encoding mechanism and pre-training capabilities, outperforms others in understanding nuanced emotional tendencies and complex contextual relationships in Chinese black humor texts. This research not only provides novel solutions for Chinese black humor text classification but also enhances our understanding of pre-trained language models in domain-specific applications.