科研社交网络中跨学科情报推荐方法研究  

A Research of Interdisciplinary Information Recommender for Scientific Social Network

作　　者：谢海涛 肖雯 黄劲松[2] Xie Haitao;Xiao Wen;Huang Jinsong(Beijing United Information Center of Science-Technology-Economy,Beijing 100044;Beijing Academy of Science and Technology,Beijing 100089)

机构地区：[1]北京科技经济信息联合中心,北京100044 [2]北京市科学技术研究院,北京100089

出　　处：《情报杂志》2019年第5期186-194,共9页Journal of Intelligence

基　　金：北京市科学技术研究院萌芽项目"深度学习技术在情报检索用户群体行为分析中的应用研究"(编号:GS201804)研究成果之一

摘　　要：[目的/意义]科研社交网络是科研人员分享学术情报和寻找合作伙伴的重要平台,对跨学科合作有极大推动作用。虽然该类平台中的推荐系统能为用户提供符合其以往偏好的学术情报,但难以对围绕特定领域形成的学术社团推荐新颖的跨学科情报。[方法/过程]将用户的差异化信息传播贡献纳入考量,设计了侧重跨学科情报推荐的协同过滤算法。通过卷积神经网络预测候选情报与学术社团偏好的匹配度,从而优化推荐结果,并克服"数据稀疏""冷启动"问题。[结果/结论]基于真实数据集进行实验,结果表明本方法对跨学科情报的推荐效果优于传统协同过滤,且推荐结果优化模型具有可扩展性。[Purpose/Significance ]Scientific research social network is an important platform for researchers to share academic information and achieve interdisciplinary cooperation. Although the recommendation system of a platform can provide academic information for users in accordance with their previous preferences, recommending novel interdisciplinary information to academic groups of specific domains is hard and rare.[Method/Process ]Taking different contributions of users during information spread into consideration, we design an interdisciplinary collaborative filtering method. Convolution neural network is used to refine candidate recommendations to match preferences of researchers. In addition, the method proposed alleviates the problems of "sparsely distributed data" and "cold start".[Result/Conclusion ]Experiments show that the method has better recommending effects than traditional collaborative filters, and has good extensibility in optimizing recommendation results.

关 键 词：科研社交网络 情报推荐 协同过滤 深度学习 推荐系统 跨学科研究 

分 类 号：G350[文化科学—情报学]