基于异构网络的企业科研合作者推荐研究

Enterprise-Oriented Researchers Recommendation Based on Heterogeneous Network

摘要:

论文研究产学研领域面向企业的科研合作者推荐问题。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；分析不同语义关系下连通企业与科研合作者的元路径；以各元路径下的路径实例为语料，利用SkipGram模型进行网络嵌入训练；采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：该方法可识别出面向企业推荐的主要有效路径，这些路径在准确率等指标上具有良好的表现；灵敏度分析也说明了该方法具有良好的鲁棒性。

Abstract:

This paper studies the problem of enterprise-oriented researchers recommendation in the field of industry-university-research. In this paper, heterogeneous networks are introduced to fuse multi-node information containing enterprises, researchers, patents and papers, as well as multi-related information such as technology context and social connections of enterprises. The paper analyzes the meta-paths connecting enterprises and researchers under different semantic relations in the network, takes the path examples under each meta-path as corpus, and train network embedding using SkipGram model, the correlation degree between nodes is expressed by vector cosine similarity, and the final recommendation list is obtained by fusing the results of different paths. The case study based on Scholarmate has demonstrated that this method can identify the main effective paths for recommending to enterprises. The paths perform well in precision and other measures. Sensitivity analysis also demonstrates that this method has good robustness.