A Tripartite Graph Recommendation Algorithm Based on Item Information and User Preference
Zhenyu Yao, Jinkuan Wang, Yinghua Han · 2019
Data sparsity and cold start are common problems in item-based collaborative ranking. To address these problems, some bipartite-graph-based algorithms are proposed, but two flaws are still involved in the proposed bipartite-graph-based algorithms. First, they cannot introduce the information of tags into recommendation model, and second, they can't effectively utilize the priorities information from users' rating data. In order to overcome the shortcomings of current bipartite-graph-based collaborative ranking methods, a tripartite-graph-based method called TGRIP is proposed in this research. In the TGRIP algorithm, a new tripartite graph is applied to introduce the information from users' rating data and make links between user layer and item layer. For the reason that users' past preference information and tag information of items have a great effect on improving the data sparsity issue and cold start problem of items, users' recommendation lists are created based on these auxiliary information. The proposed algorithm keeps the items most likely to be accessed by a target user at the top of recommendation result. This means these items come under top-n recommendation list, which are most likely chosen by target users. Experiments show that the proposed approach achieves a remarkable improvement compared to bipartite-graph-based recommendation algorithm in terms of recall, precision, F1-score and normalized discounted cumulative gain metrics.