Research on aspect-based sentiment analysis of homestay online comments based on BERT-BiLSTM-Multi-Head Attention
Yang Yang, Fang Liu · 2025
In the digital age, online comments on travel social platforms are gradually becoming more diversified. The conventional aspect⁃based sentiment analysis model can not well represent the deep⁃level attribute features of tourism products. In order to more effectively mine and accurately capture fine-grained emotional information within the internal entities of information, a sentiment analysis model based on BERT-BiLSTM-Multi-Head Attention is proposed. The model uses BERT pre-trained language model for text representation, BiLSTM sequence modeling, Multi-Head Attention mechanism Introduced. The experiment was carried out on datasets such as online comments, Weibo texts and homestay comments. The experimental results show that Multimodal Aspect-based Sentiment Analysis model has improved significantly, and the classification accuracy is 89.6%, which is better than other baseline models. Ablation experiments are conducted, which effectively verified the reliability of each module in the MABSA model for expressing the fine-grained sentiment features of text sequences.