Consumer purchase behavior prediction based on user-generated content sentiment analysis algorithm

Abstract

With the increasing socialization of e-commerce platforms, user-generated content (UGC), such as short video reviews and image-based evaluations, has become a core factor influencing consumer decisions. However, existing prediction models often suffer from the following bottlenecks: the semantic gap between text and image sentiment features; and neglecting the temporal dynamic impact of emotional fluctuations on individual purchasing decisions. This paper proposes a multimodal sentiment-enhanced behavioral graph attention network model (MS-BGAN). This model aligns multidimensional sentiment in UGC through feature-level fusion technology and constructs a novel sentiment-weighted heterogeneous graph, utilizing an improved graph attention mechanism to capture the social traction between consumers and the attractiveness of products. Experimental results show that on real datasets from Amazon and TikTokShop, MSBGAN improves the prediction accuracy (ACC) by 12.4% compared to the baseline model, providing a new theoretical framework for precision marketing.