摘要
【目的/意义】通过网络表示学习方法学习和发现作者间的关联性,提高推荐准确率,更好地进行关联推荐。【方法/过程】基于2010-2017年中国管理科学领域的数据构建基于网络表示学习的作者合作推荐模型,通过多关系映射获取到的多关系序列作为语料库,采用Word2vec方法进行网络表示学习训练,并利用余弦相似度方法计算作者间相似度。【结果/结论】本文算法推荐精度达到82.8%,其推荐精度显著提高;融合元路径(元结构)特征的推荐精度大幅提高,能为作者合作伙伴的选择提供建议和参考,对实践具有一定的指导意义。
【Purpose/significance】By learning and discovering the association between authors through the network representation learning method,the accuracy and efficiency of recommendation can be improved,and the association recommendation can be better carried out.【Method/process】Based on the information from 2010 to 2017 in the field of Management Science in China,a collaborative recommendation model of authors based on network representation learning is constructed.A heterogeneous relationship traversal algorithm based on meta-path and meta-structure is used to obtain the relationship sequence between authors by multi-relational mapping.Using relational sequences as corpus,the network embedding training is carried out by using Word2 vec method and using cosine similarity to calculate the similarity between authors.【Result/conclusion】The recommendation accuracy of the proposed algorithm is 82.8%,and its recommendation accuracy is significantly improved;the recommendation accuracy of integrating multiple meta-paths and meta-structural features is greatly improved,it can provide suggestions and references for the selection of partners,and has a certain guiding significance for practice.
作者
刘云枫
孙平
葛志远
LIU Yun-feng;SUN Ping;GE Zhi-yuan(School of Economics and Management,Beijing University of Technology,Beijing 100124,China)
出处
《情报科学》
CSSCI
北大核心
2020年第2期75-80,共6页
Information Science
基金
国家自然科学基金面上项目“基于类比推理的短生命周期无形体验品需求预测”(71672004)
北京市自然科学基金资助项目“多重共现耦合的科技知识网络关联发现研究:链路预测的视角”(9174029).
关键词
作者合作推荐
网络表示学习
异构信息网络
推荐
author cooperative recommendation
network embedding
heterogeneous information network
recommendation
