To alleviate data sparsity and cold-start problems of traditional recommender systems (RSs), incorporating knowledge graphs (KGs) to supplement auxiliary information has attracted considerable attention recently. Since the construction of these KGs is independent of the collection of historical user-item interactions, information in these KGs may not always be helpful to all users. Simply integrating KGs in current KG-based RS models does is not guaranteed to improve recommendation performance. In this talk, we discuss, we discuss our proposal of a novel knowledge-aware recommendation model (CG-KGR) that enables ample and coherent learning of KGs and user-item interactions. Specifically, CG-KGR first encapsulates historical interactions to interactive information summarization. Then, CG-KGR utilizes it as guidance to extract information out of KGs, which eventually provides more precise personalized recommendation.
Learn more about the 2021 Microsoft Research Summit: https://Aka.ms/researchsummit