Intelligent recommender system may trap users in phenomena such as "information cocoons," "filter bubbles," and "echo chambers," leading to over-specialization. Individual psychological traits are internal representations of external behavior. Exploring and validating the potential role of online users' psychographics in mitigating their over-specialization tendencies，holds theoretical and practical significance for improving and evaluating of user-centered recommender system. Using natural language processing and machine learning experimental methods, based on psychographic segmentation theory, this study investigates the relationship between users' demand for recommendation diversity (i.e., "specialization" tendency) and their psychological traits through opinion mining and psychological computation methods. It then integrates psychographics into user preference models and validates them through large-scale preference behavior data and recommendation algorithm experiments. The study finds that: psychographics can effectively explain and predict users' demand for recommendation diversity. By integrating user preference models, psychographics can mitigate over-specialization while ensuring recommendation accuracy. Compared to personality traits, values have a relative advantage in explaining and mitigating users' over-specialization tendencies.