基于异构网络的企业科研合作者推荐研究  

Research on Recommendation of Enterprise Scientific Research Collaborators Based on Heterogeneous Network

作　　者：杨娜 刘钱 余小菊 Yang Na;Liu Qian;Yu Xiaoju(International Business School,Shaanxi Normal University,Xi'an 710119,China)

机构地区：[1]陕西师范大学国际商学院,陕西西安710119

出　　处：《科技管理研究》2024年第14期234-242,共9页Science and Technology Management Research

基　　金：国家自然科学基金项目“群组推荐方法及其应用研究:基于偏好冲突与交互影响分析视角”(72102134);教育部人文社会科学基金项目“考虑群组成员间偏好冲突与社会影响的组推荐方法及其应用研究”(20YJC630189);陕西省自然科学基础研究计划项目“基于群组偏好建模的组推荐方法及其应用研究”(2021JQ-315)。

摘　　要：旨在研究产学研领域中面向企业的科研合作者推荐问题,以改进现有方法中仅使用专利、合作关系等单一信息的现状,以及避免在可移植性方面的局限性。提出基于异构网络向企业推荐潜在科研合作人员的方法:首先引入异构网络,融合企业、科研人员、专利和论文等多元节点信息,以及企业技术需求和社交关联等多元关联信息;其次分析不同语义关系下连通企业与科研合作者的元路径,并以各元路径下的路径实例为语料,运用SkipGram模型进行网络嵌入训练,用向量余弦相似度表示节点之间的关联程度;最后融合不同路径下的推荐结果,得到最终的科研合作者推荐列表。基于Scholarmate的实例验证表明,元路径1和路径3的推荐效果最好,而综合各条元路径时模型在准确率和特异度指标上表现更好;此外,在符合企业实际情况的不同推荐列表长度下,模型各指标变化不大且处于理想水平,且综合多条元路径时模型的鲁棒性更强。此方法可为企业解决科技人才获取难的问题提供解决方案,并为企业的技术需求分析和产学研领域的社交关系分析提供思路参考。This paper aims to study the recommendation problems of scientific research collaborators for enterprises in the industry-university-research field to improve the status of existing methods using only single information such as patents,partnerships,and relationships,and to avoid limitations in portability.The method of recommending potential scientific research partners to enterprises based on heterogeneous network is proposed:heterogeneous networks are introduced to integrate multiple node information such as enterprises,researchers,patents and papers,as well as multiple association information such as enterprise technical requirements and social connections;the meta-paths connecting enterprises and research collaborators under different semantic relationships are analyzed;using the path instances under each element path as corpus,the SkipGram model is used for network embedding training;vector cosine similarity is used to represent the degree of correlation between nodes;finally,the recommendation list of scientific research collaborators is obtained by integrating the recommendation results under different paths.Example verification based on Scholarmate shows that meta-paths 1 and meta-paths 3 recommend the best,while the integrated model performs better in accuracy and specificity index.In addition,under the length of different recommendation lists in line with the actual situation of the enterprise,the indicators of the model change little and are at the ideal level,and the model is more robust when integrating multiple meta-paths.This method can provide solutions for enterprises to solve the problem of obtaining scientific and technological talents,and provide reference for the analysis of technical needs and social relationship analysis in the field of industry,university and research.

关 键 词：科研合作者推荐 潜在科研合作人员 异构网络 元路径 网络嵌入 Skig-Gram模型 

分 类 号：G203[文化科学—传播学] G302