Personalized literature recommendation based on heterogeneous entity academic network

Abstract

Although researchers have benefited from big scholarly data, it is still very difficult for them to quickly and accurately find the suitable literature in the massive literature. In recent years, the research on personalized literature recommendation with the help of academic data has gradually attracted the attention of scholars. However, the existing works are mainly based on the similarity of literature content, and ignore the important information of scholars such as research fields and affiliations, which leads to insufficient personalization of recommendation results and there is still room for improvement in accuracy. In this paper, a new personalized literature recommendation method named PR-HeAN is proposed. This method also considers the similarity of literature, and more importantly, introduces a heterogeneous entity academic network. It combines the literature similarity obtained from a K-order literature co-citation network with the literature recommendation probability obtained from the heterogeneous entity academic network to generate the recommendation results. The heterogeneous network is generated from the K-order literature co-citation network, which integrates five types of academic entities, including literature, scholars, research fields, affiliations and publication venues, and enriches the representation information of the literature. The experimental results on two datasets show that the proposed method outperforms four baseline algorithms in terms of recall, accuracy and F1 value.

Keywords

• Personalized literature recommendation
• Heterogeneous entities
• Academic network
• Graph Convolutional Network