Graph-embedding-inspired article recommendation model

Intelligent recommendation systems(IRSs) need to face the problems of sparse data and cold start. As one of the intelligent recommendation scenarios, article recommendation has rich and various article information, which is a significant difference from other scenarios. However, the existing methods of article recommender do not fully consider the article information or ignore the implicit correlation between users and articles. Therefore, this paper proposes a Graph-Embedding-inspired Article Recommendation Model (GE-ARM): we capture the User -Article correlation feature embeddings by constructing a Bipartite Graph with Resource Allocation (RA-BG); considering the different contributions of the Title, Abstract and Tag, Citation, design an Attention-based Dual-AutoEncoder, combining with Collaborative Filtering to fully capture the article feature embeddings; and update the correlation and article feature embeddings with Probabilistic Matrix Factorization to perform the final recommendation task. The experimental results highlight that GE-ARM outperforms the other known methods on four datasets.

Automated summary

This paper proposes a Graph-Embedding-inspired Article Recommendation Model(GE-ARM) that captures the user-article correlation feature embeddings by constructing a Bipartite Graph and updates the correlation and article feature embeddings using Probabilistic Matrix Factorization. Experimental results demonstrate that GE-ARM outperforms other methods on four datasets.