Enhanced graph recommendation with heterogeneous auxiliary information

Abstract The boom in the field of movies and TV programs, which is a kind information overload, may lead to poor user experience are detrimental healthy development industry, hence personalized program recommendation crucial. Since names, labels, synopsis highly condensed languages, enable better semantic representations for recommendations enrich completeness requirements data resources, we propose an enhanced graph with heterogeneous auxiliary (EGR-HA), focusing on knowledge representations, neural network-based node updates. Firstly, multi-source fused supplement semantics obtain initial that contain rich semantics, then embedding aggregated multiple layers through networks model higher-order interaction history realize representation update; finally, viewing prediction performed based deep recommendation. final experiment results indicators, such as normalized discounted cumulative gain (NDCG), hit rate (HR) root mean square error (RMSE), verified effectiveness this method by comparing various methods.