基于模体挖掘面向在线社交网络中虚拟社区的群推荐系统研究

A group recommender systems (GRSs) is an information system which suggest items for meeting the common interests of a group of people engaged in a group activity. Though there have been some studies on group recommender systems, they focus on off-line environments and established groups. These days, many group activities and interactions are done in a virtual space, so research on GRSs for virtual communities in online social networks will become a hot issue in the recommender systems field. To support group activities in online communities, this proposal will research on the GRSs for virtual communities in online social networks based on the motif discovery. We will use some theories and methods such as social network analysis, intelligent optimization method and group decision theory et cetera to solve the relative problems. In the research process, we will first study the modeling method of online social network based on Social-Affiliation Network, and then we will concern the big motifs' and virtual communities' discovery in the Social-Affiliation Network. A novel method called "Motif Percolation Method" will be proposed in the process of discovery of virtual communities. Next, we will research choice effect and social influence effect in the process of community growth by tracking the changes of both social relationships and behaviors of the community members. To the different type of communities, we will research the strategies for aggregating individuals' preferences, generating recommendations to the group and coordinating and optimizing the recommendations in the group. Finally, to check the effect of the theories, methods and strategies mentioned above, we will develop a prototype of group recommender system that suggests the Baidu Apps to the group in the virtual communities of Baidu Paste Bar. These researches contribute to understand the formation of group preferences and group behavior patterns in online social networks and have important theoretical significance or great practical value for improving marketing efficiency and enhancing marketing efforts of social networking platform through establishing the group recommender systems for virtual communities and taking social marketing activities.

群推荐系统是为参与共同活动的群体推荐满足群中所有成员的共同爱好的信息系统，现有成果主要支持线下已成团群体。随着在线社交网络的出现，具有相似兴趣的群体形成了各类虚拟社区，面向社交网络中虚拟社区的群推荐系统将成为推荐领域的热点问题。本项目将应用社交网络分析方法、智能优化方法以及群决策相关理论，研究基于社会归属网络的社交网络表示模型；研究社会归属网络中的大模体发现方法以及基于模体渗透法的虚拟社区发现方法；通过跟踪社区成员的社会联系与行为变化，研究虚拟社区形成过程中选择和社会影响效应的不同影响，并针对影响程度不同的虚拟社区，研究群偏好集结与模型优化策略、群推荐列表产生方法和协调策略；通过面向百度贴吧群推荐百度APP以验证成果的有效性。本项目研究有助于理解社交网络环境下群体的偏好形成和行为模式，对于构建面向虚拟社区的群推荐系统和开展社会化营销，提高社交网络平台的营销效率和效果具有理论意义和实际价值。

本项目面向在线社交网络中虚拟社区的群推荐问题，应用社交网络分析、机器学习、智能优化以及群决策理论等相关方法，围绕社会归属网络中的社交网络表示模型与个体成员偏好获取方法、社会归属网络中的大模体发现方法、基于模体渗透的虚拟社区发现方法、群偏好集结与模型优化策略、群推荐列表产生方法和协调策略等开展了系统的研究。针对社会归属网络中的节点重要性问题，项目组提出一种基于节点间电导的算法计算个体节点权重；在个体成员偏好获取方面，项目组提出了一种概率主题模型预测在线社交网络中用户的人格特质，该成果为在线社交网络中用户建模提供了新的思路；针对抱怨群体社区的发现难题，项目组还提出了一种基于熵权重模型的在线抱怨主题影响力度量模型，并探讨了基于静态和动态特征的抱怨主题影响力的度量方法，从而挖掘出抱怨群体社区；针对在线社交网络中虚拟社区的群推荐问题，项目组提出了一种基于改进的协同过滤与中位数加权平均的群推荐方法；项目组还将群推荐问题建模为一个双向过程，提出了一种用于群推荐的双向张量分解模型，以刻画个体内在偏好和群体影响间的交互作用。本项目研究有助于理解社交网络环境下群体的偏好形成和行为模式，对于构建面向虚拟社区的群推荐系统和开展社会化营销，提高社交网络平台的营销效率和效果具有理论意义和实际价值。项目组在研究期内先后在《International Journal of Production Economics》等国内外学术期刊和会议上共发表论文 34篇，其中被SCI收录10篇，EI收录4篇；承办了第九届国际信息管理中国夏季研讨会和第十届中国管理学年会，组织和参加国内外合作与学术交流有20余次；获得2015年安徽省科技进步奖二等奖1项，2015年教育部高校科研优秀成果奖1项；支持了5名博士生和7名硕士生的学位论文。