This study proposes a novel video recommendation approach that leverages implicit user feedback in the form of viewing percentages and social network analysis techniques. By constructing a video similarity network based on user viewing patterns and computing centrality measures, the methodology identifies important and well-connected videos. Modularity analysis is then used to cluster closely related videos, forming the basis for personalized recommendations. For each user, candidate videos are selected from the cluster containing their preferred items and ranked using an ego-centric index that measures proximity to the user's likes and dislikes. The proposed approach was evaluated on real user data from an Asian video-on-demand platform. Offline experiments demonstrated improved accuracy compared to conventional methods such as Naive Bayes, SVM, decision trees, and nearest neighbor algorithms. An online user study further validated the effectiveness of the recommendations, with significant increases observed in click-through rate, view completion rate, and user satisfaction scores relative to the platform's existing system. These results underscore the value of incorporating implicit feedback and social network analysis for video recommendations. The key contributions of this research include a novel video recommendation framework that integrates implicit user data and social network analysis, the use of centrality measures and modularity-based clustering, an ego-centric ranking approach, and rigorous offline and online evaluation demonstrating superior performance compared to existing techniques. This study opens new avenues for enhancing video recommendations and user engagement in VOD platforms.
翻译:暂无翻译