Research on User Dynamic Preference Modeling and Recommendation System Based on Text Sentiment Transfer

User preferences in recommendation systems are often dynamic and influenced by contextual factors, among which textual sentiment plays a pivotal role. However, conventional recommendation models frequently assume static user interests, failing to capture the evolving nature of emotional expressions in textual interactions. This paper proposes a novel approach that integrates text sentiment transfer mechanisms to model user dynamic preferences effectively. By leveraging a domain-adaptive sentiment encoder combined with a recurrent attention-based preference tracker, we are able to identify latent shifts in user sentiment and incorporate them into real-time recommendation strategies. Experiments conducted on multiple real-world datasets demonstrate the superiority of our method in capturing finegrained user preferences and improving recommendation accuracy. The results validate that incorporating sentimentaware dynamics significantly enhances user satisfaction and recommendation diversity. This research contributes a new perspective on sentiment-driven preference modeling, offering promising directions for future adaptive recommendation systems.