Research on Emotion Classification Method of Bullet Screen Text Based on Video Topic Discovery

With the rapid development of short video platforms, bullet screens (danmu), as a real-time way for users to express their emotions, contain rich emotional information and interactivity. However, the brevity and colloquial nature of bullet screen texts present certain challenges for emotion classification. To address this, this paper proposes an emotion classification method for bullet screen texts based on the BERT_BiLSTM_CNN model, combined with video topic features extracted by the improved HW-LDA model. The experimental results demonstrate that the emotion classification model, which incorporates topic features, significantly outperforms models without topic features in terms of accuracy, recall, and F1 score. This study highlights the importance of topic features in bullet screen emotion classification and provides new insights for emotion analysis of short texts.