Scholarly Paper Recommendation via Related Path Analysis in Knowledge Graph

Abstract

Recommending helpful and interesting scholarly papers for researchers from a large number of scholarly papers is the main way to improve research efficiency. Traditional collaborative filtering or content-based recommendation methods do not have a better-fused knowledge graph and have method bottlenecks such as cold start and poor interpretation. Based on the knowledge-aware path recurrent network (KPRN), this paper proposes a method for recommending scholarly papers that combines user preferences and knowledge graph path information. Firstly, a delayed extension bi-directional breadth-first search path algorithm is proposed to find the path between two nodes in the knowledge graph with low time complexity. Then, the user preference vector is generated by the user's historical paper operation. Finally, the LSTM cyclic neural network model is used to extract the information of multiple paths and combine it with user preferences to obtain the list of recommended papers. The experimental results show the validity and good interpretability of this method.