软件导刊2026，Vol.25Issue(8)：15-22,8.DOI:10.11907/rjdk.251859
基于学者属性与多元路径信息的科研合作预测
Scientific Collaboration Prediction Based on Researcher Attributes and Multiple Meta-Path Information
摘要
Abstract
Academic collaboration broadens research horizons and promotes diversity and innovation in scientific research.In recent years,fu-sion methods based on graph representation learning in heterogeneous academic networks have demonstrated good performance.However,ex-isting collaborative prediction studies often utilize only a small amount of meta-path information and lack the introduction of other types of in-formation,resulting in suboptimal prediction results in related category scenarios.Introducing overall category information can comprehensive-ly capture different collaboration motivations.To handle overall category information,a novel hierarchical meta-path fusion graph neural net-work model,HMFG,is proposed.This model models from both scholar individual attributes and heterogeneous structural information,obtains neighbor features through lightweight mean aggregation,unifies meta-path dimensions using multi-layer feature projection,and combines multi-head attention and gating mechanisms to achieve high-quality semantic fusion.Experiments conducted on the WOS and SCHOLAT data-sets show that HMFG exhibits stable and excellent predictive performance on both datasets.In the WOS dataset constructed based on text fea-tures,ACC,F1,and AUC achieve minimum performance improvements of 2.29％,2.04％,and 1.08％,respectively.关键词
科研合作预测/异构学术网络/图表示学习/链接预测Key words
academic collaboration prediction/heterogeneous academic network/graph representation learning/link prediction分类
信息技术与安全科学引用本文复制引用
马健兵,张同强,黄博..基于学者属性与多元路径信息的科研合作预测[J].软件导刊,2026,25(8):15-22,8.基金项目
四川省科技计划项目(2023YFS0192) （2023YFS0192）
