Research on Sentimental Tendency Classification of Chinese Internet Memes Based on Self-Learning Decision Level Fusion Model
Peng Nan, Nuo Qun, Wen Yao, Yan Hou, Jin Yang, Jian Dong · 2022
The analysis of the sentimental tendency of Internet memes has broad practical application prospects in politics, economy, society, humanities and other fields. The research on classifying Internet memes using artificial intelligence methods has attracted more and more researchers' attention. As a multimodal expression, Internet memes can be divided into two parts: images and texts embedded in images. Currently, the main research object in this field is the Internet meme of English text. This paper establishes a data set consisting of 8000 Chinese Internet meme pictures, and a self-learning decision level multimodal fusion model based on Bert and resnet50 network models is proposed to classify the data set into three categories according to sentimental tendency: positive, neutral and negative. The model's accuracy is 0.79, and the F1 score is 0.76. The above experiments preliminarily verify the feasibility of using artificial intelligence methods to classify Chinese Internet memes. In addition, this paper also proposes a new evaluation index, which is more suitable for the effect evaluation of non-discrete classification tasks such as sentimental tendency classification.