Continual knowledge base enhancement for communal recommendation using homogeneous opinion-dynamics in heterogeneous social-networks

User decision-making in online social networks is deeply shaped by interpersonal influence, yet conventional social recommendation models often rely on static neighbor aggregation or global trust propagation, limiting their adaptability to continually evolving user behaviors. To address this, we propose CROS — Continual knowledge base enhancement for communal Recommendation using homogeneous Opinion-dynamics in heterogeneous Social-networks. CROS quantifies user influence through structural roles in heterogeneous networks, integrating both cross-network social relations and community-level opinion dynamics. Moreover, it incrementally enhances the knowledge graph with multi-source, time-evolving data and models user interactions via evolutionary game theory, linking opinion formation to preference evolution. Experiments on three real-world datasets – Amazon, Ciao, and Epinions – demonstrate that CROS consistently outperforms state-of-the-art baselines, by achieving up to 5% higher NDCG and 3% lower MAE or RMSE. Overall, CROS delivers more accurate, robust, and adaptive recommendations under dynamic, sparse, and cold-start conditions.