Research and Application of Personalized Recommendation Based on Knowledge Graph.

Text data resources in the power domain have become increasingly abundant in recent years with the large scale popularization of information office in the power sector, but workers are facing an increasingly severe problem of data information overload. Since the concept of knowledge graph have been proposed, researchers have used professional datasets in various fields to construct corresponding knowledge graphs and proposed various knowledge graph completion algorithms to solve the problem of missing entity and relation links. In this paper, we introduce the knowledge graph as auxiliary information into the recommendation system of power domain. Our method uses translation-based models to learn the representations of users and items and applies them to optimize the recommender system. In addition, to address users diverse interests, we also build user profiles in our method to aggregate a users history with respect to candidate items. According to the characteristics of the data and the representativeness and universality of the data, extensive experiments are conducted on the Citeulike. We apply our approach to the power domain and construct the knowledge graph of the power domain dataset. The results validate the effectiveness of our approach on recommendation.