News Sequence Recommendation Model with Dual-View Category Enhancement

Personalized news recommendation enriches readers’ news consumption experience by aligning news articles with their preferences. However, existing methods often fall short in extracting complete news topic features and accommodating variations in reader preferences. To tackle these challenges, a dual-view category-enhanced news sequence recommendation model is proposed. In the news representation module, we enhance CNN-derived topic vectors with a context-weighted word-based approach, rectifying dispersed topic extraction. It yields a comprehensive news representation vector. In the user representation module, a graph construction method, utilizing a sliding window technique, captures temporal user-centric features and multiple user interests. Dual-view learning ensures both temporal and category-based features are encoded effectively. Experiment results on real-world datasets indicate that the proposed model exhibits exceptional news recommendation performance, with significantly higher quality compared to the comparison models.