软件导刊2026，Vol.25Issue(8)：40-48,9.DOI:10.11907/rjdk.261291
大学生学研产匹配的LLM-GNN路径预测
LLM-GNN-Based Path Prediction for Student Academic-Research-Industry Matching
摘要
Abstract
To address fragmented information,insufficient semantic understanding,limited use of relational structure,and weak path-level explanations in the transition from coursework and research training to industrial positions,this study proposes an academic-research-indus-try path prediction method that combines large language models with graph neural networks.Students,research projects,enterprise positions,courses,and skills are organized into a tripartite heterogeneous graph.Large language models extract skill entities,experience evidence,and semantic vectors,while graph convolutional networks aggregate multi-hop neighborhood information.Gated post-fusion and dynamic hard neg-ative sampling are then used to predict student-project and student-position links.High-scoring candidate links are further organized into growth paths toward target positions,and maximum a posteriori inference,prerequisite consistency,and difficulty smoothness are adopted to quantify path feasibility and reasonableness.Transfer validation on the SCHOLAT academic social network achieves 98.72％accuracy,0.9934 average precision,0.9929 ROC AUC,and 93.75％positive recall.Ablation results show that LLM semantic encoding,GCN topology aggrega-tion,gated fusion,and dynamic negative sampling each make independent contributions to predictive performance and path reasonableness.The proposed LLM-GNN framework provides interpretable decision support for university-industry integration,research project selection,and employment guidance,while real campus deployment still requires authorization,de-identification,ethics review,and staged empirical validation.关键词
大语言模型/图神经网络/链接预测/产教融合/教育数据挖掘Key words
large language model/graph neural network/link prediction/industry-education integration/educational data mining分类
信息技术与安全科学引用本文复制引用
孙恒一,黄雨涵,余亮,张笑晨,罗洛阳,汤庸..大学生学研产匹配的LLM-GNN路径预测[J].软件导刊,2026,25(8):40-48,9.基金项目
广东省教育评估协会课题(GDPG25064) （GDPG25064）
广东省普通高校重点领域专项项目(2025ZDZX1047) （2025ZDZX1047）
广东省重点建设学科科研能力提升项目(2025ZDJS065) （2025ZDJS065）
