Sentiment Analysis Model for Live Streaming Comments Based on Bi-LSTM and Attention Mechanism

With the rapid development of live streaming, user comments in live rooms have become an important medium reflecting content quality and audience sentiment. To address the characteristics of live bullet-screen comments, such as temporal dependency, sparsity, and implicit sentiment, this paper proposes a joint modeling approach based on Bidirectional Long Short-Term Memory (Bi-LSTM) and Attention Mechanism (Bi-LSTM+Attention). Taking Douyin (TikTok) live streaming platform as the research subject, an automated data collection system was built using Python to achieve structured crawling of comments from specific live rooms. A high-quality corpus was constructed through text cleaning, denoising, and annotation preprocessing. This paper presents a sentiment analysis model for live streaming comments based on Bi-LSTM and Attention mechanism. The model leverages Bi-LSTM to capture contextual information in comments and employs attention mechanisms to highlight key sentiment words, thereby improving the accuracy of sentiment classification. Experimental results demonstrate that, compared to traditional sentiment analysis methods, the proposed model achieves higher classification accuracy on live comment datasets and effectively identifies emotional tendencies in user comments.