Embedding ranking-oriented recommender system graphs

Graph-based recommender systems (GRSs) analyze the structural information available in graphical representation of data to make better recommendations, especially when direct user-item relation is sparse. Ranking-oriented GRSs mostly use preference (or rank) for measuring node similarities, from which they can infer recommendations using neighborhood-based methods. In this paper, we propose PGRec, a novel model-based ranking-oriented recommendation framework. Unlike many other graph-based methods, PGRec extracts vector representations users and preferences graph structure called PrefGraph, models entity relations, feedbacks, content. A general graph-embedding process improved applied extract entities. The resulting embeddings are then used predicting target user’s unknown pairwise by neural network based on list generated user. We have evaluated proposed method’s performance against state art algorithms. Our experiments show that outperforms baseline algorithms terms NDCG metric several datasets.