Research on Sentiment Analysis Method Based on the Collaborative Modeling of Attention Mechanism and Sentiment Lexicon

Traditional sentiment lexicon methods, relying on static semantic annotations, struggle to capture dynamic emotions in context (e.g., the negative modification in “unhappy”). In contrast, purely data-driven attention mechanisms, lacking prior semantic guidance, are highly susceptible to noise interference. Addressing the challenges of complex emotion expressions in social media texts and the inability to effectively capture users' personalized needs, this paper proposes a sentiment analysis method via collaborative modeling of the attention mechanism and sentiment lexicon. The approach embeds the semantic polarity and intensity of sentiment lexicons into the word vector space, dynamically calculating contextual association weights through the attention mechanism to form a bidirectional “semantic prior guidance - dynamic feature focusing” interaction model. This significantly enhances emotion recognition accuracy in complex contexts. Additionally, by integrating users' historical behavior data, personalized emotion preference modeling is performed to enable fine-grained analysis of social media texts. Experimental results demonstrate that the proposed method substantially improves classification accuracy in social media emotion recognition tasks, offering robust technical support for applications like content recommendation and public opinion monitoring.