Group Emotion Recognition for Weibo Topics Based on BERT with TextCNN
Donghong Shan, Huili Li · American Journal of Information Science and Technology · 2023
Social media platforms, including Weibo, have become an integral part of people's daily lives, where users engage in discussions, share opinions, and express their emotions regarding trending topics. However, as the volume of information and content continues to increase, individuals face challenges in accessing relevant information. To address this issue, sentiment analysis has been employed in this study to focus on group sentiment identification for Weibo topics. Due to the potential involvement of multiple sentiment categories in Weibo topics, the main algorithm used in this research combines BERT and TextCNN for text multi-label classification. This approach aims to predict the possible collective emotional reactions of the public. Macro-F1 has been chosen as the evaluation criterion, with the baseline algorithm achieving a score of 0.3339, while our model achieved a slightly improved score of 0.3514. This improvement demonstrates the efficacy of the proposed algorithm. This paper makes full use of the self-attentive mechanism of BERT combined with the convolutional layer and pooling operation of TextCNN to extract local features. The generalization ability and sentiment classification accuracy of the model are improved. The results of text multi-label classification for group sentiment recognition of microblog topics demonstrate the superiority of the model algorithm in this paper. This study carries significant implications for understanding the public's emotional responses to popular topics on social media. It provides valuable insights for further exploration and advancement in the field of sentiment analysis within the realm of social media.