Hypergraph-based Academic Paper Recommendation

EasyChair Preprint 8541

Abstract

Academic paper recommendation aims to provide personalized recommendation services for scholars from massive academic papers. Deep Learning-based Collaborative Filtering plays an important role in it, and most of existing method are based on bipartite graph, which causes it fail to realize multi-features fusion, and the over-smooth property of GCN limits the generation of embedding with high-order similarity, resulting in the decline of recommendation quality. In this paper, we propose a hypergraph-based academic paper recommendation method. Based on hypergraph, APRHG (Academic Paper Relation HyperGraph) is constructed to not only model the complex academic relationship between users and papers, but also realize the multi-features fusion. In addition, the L-HGCF (Light HyperGraph based Collaborative Filtering) algorithm, which could mine high-order similarity between papers, is proposed to provide trusted recommendations. We conduct experiments on the public dataset, and compare the performance with several deep learning based Collaborative Filtering to confirm the superiority of our method. In addition, we conduct detailed ablation experiments to verify the rationality of the components and hyper parameter design of our proposed L-HGCF algorithm.

Keyphrases: academic paper recommendation, collaborative filtering, hypergraph

@booklet{EasyChair:8541,
  author    = {Yu Jie and He Junchen and Xu Lingyu},
  title     = {Hypergraph-based Academic Paper Recommendation},
  howpublished = {EasyChair Preprint 8541},
  year      = {EasyChair, 2022}}