Academic Paper Recommendation System Based On Heterogeneous Graph

Academic paper is an important achievement and the main carrier for the researchers to conduct scientific research.To tackle the massive academic papers in the database,there exist lots of research works about academic paper recommendation by utilizing relevant information such as citation and content.However,as to the inherent character of academic paper with heterogeneous feature representation,most of the recommendation systems only consider the superficial information of citation and content,regardless of the dependencies between them.As a consequence,research and analysis on academic paper representation and similarity learning have been conducted,and get some major results described as follows.First,as to the issue of various feature type representation,the feature representation method of academic papers based on heterogeneous graph is proposed.This method is able to express the dependencies between heterogeneous information of academic paper.Second,a graph-based framework in a semi-supervised manner is utilized to propose the similarity learning method based on heterogeneous graph.Experiments on AAN show that the proposed method outperforms traditional methods.Third,a heterogeneous graph-based academic paper recommendation system is designed and implemented based on the research above,which is capable of recommending relevant papers for the researchers according to the input paper efficiently and effectively.