现代情报2025，Vol.45Issue(7)：26-35,10.DOI:10.3969/j.issn.1008-0821.2025.07.003
融合多维学术特征的引文推荐:一种基于异质图卷积网络的方法
Citation Recommendation Integrating Multidimensional Academic Features:A Method Based on Heterogeneous Graph Convolutional Networks
摘要
Abstract
[Purpose/Significance]Most existing citation recommendation methods adopt meta-path-based network representation learning approaches,which often overlook complex interactions between nodes and heavily rely on domain knowledge.[Method/Process]This paper proposed a heterogeneous graph convolution network-based method to effectively integrate multi-dimensional academic features for improving recommendation accuracy.This proposed method first used pre-trained BERT models to extract semantic features from papers.Then,an attention-aware graph convolutional neural network was designed to automatically learn neighborhood information of nodes in the heterogeneous academic information network.Finally,this method combined the network topology and semantic information to generate paper representations.[Result/Conclusion]Extensive experiments on three datasets demonstrate that the proposed method outperforms baseline models on all evaluation metrics.Case studies further indicate the effectiveness and applicability of the proposed method in the task of citation recommendation.关键词
引文推荐/图卷积网络/异质信息网络/注意力机制/自然语言处理Key words
citation recommendation/graph convolutional network/heterogeneous information network/attention mechanisms/natural language processing分类
信息技术与安全科学引用本文复制引用
柳亚,朱莉,毛谦昂,王佳鑫,颜嘉麒,陈曦..融合多维学术特征的引文推荐:一种基于异质图卷积网络的方法[J].现代情报,2025,45(7):26-35,10.基金项目
国家社会科学基金项目(项目编号:21BGL223) （项目编号:21BGL223）
国家自然科学基金项目(项目编号:72171115、72071104). （项目编号:72171115、72071104）
