A Concept Knowledge Base For Recommendation System

In this paper, we introduce a concept knowledge-base which is used forcross-domain recommendation system. This knowledge-base could helpthe computer have an association mind just like association in humanmind. The knowledge-base could answer the question like “Which personis in your mind when we talk about ZhuGe Liang(诸葛亮)”. The Conceptknowledge-base is built upon three Chinese online encyclopedia-baidubaike, hudong baike and Chinese encyclopedia. After data integration,segmentation and annotation, synonym merging, concept co-occurrencecalculating and association relationship computing, we generate theassociation knowledge-base. To our whole knowledge, this is the firstChinese association knowledge-base which provides the associationrelationship between two Chinese terms. User study showcases that ourassociation knowledge-base has the capability of association beyond theexpectation of human users. One important application of our associationknowledge-base is in Cross-Domain recommendation system. In thiscross-domain recommendation system, we could not only recommendproduct within this domain but also product beyond the domain boundary.With the help of our association knowledge-base, both the novelty, coverage, and the capability of explanation were improved. Along withthe cross-domain recommendation, our association knowledge-base couldbe applied in query completion, query rewriting, text mining and texttopic inference applications. Our association knowledge-base can also beused to help construct other Chinese knowledge-base or help the queryanswering in other knowledge-base. We hope our Chinese associationknowledge-base could help both Chinese AI application and specificallyChinese text AI task.