基于事件的社交网(event-based social networks,EBSN)中的个性化推荐服务是一个十分重要且颇具应用价值的问题,现有研究工作主要基于普通图来对EBSN中的关系进行建模,但由于EBSN是一种异构型复杂社交网络,具有多种不同类型实体,因而用...基于事件的社交网(event-based social networks,EBSN)中的个性化推荐服务是一个十分重要且颇具应用价值的问题,现有研究工作主要基于普通图来对EBSN中的关系进行建模,但由于EBSN是一种异构型复杂社交网络,具有多种不同类型实体,因而用普通图建模EBSN会存在高维信息丢失问题,导致推荐质量降低.基于此,首先提出一种基于超图模型的EBSN个性化推荐(hypergraph-based personalized recommendation in EBSN,PRH)算法,其基本思想在于利用超图具有不丢失高维数据信息之特点来更准确地对EBSN中复杂社交关系数据进行高维建模,并利用流形排序正则化计算获取初步推荐结果.其次,又分别从查询向量设置方式改进和对不同类超边施以不同权重等角度,提出了优化的PRH(optimized PRH,oPRH)算法以进一步优化PRH算法所获推荐结果,从而实现精准推荐.扩展实验表明,基于超图的EBSN个性化推荐及其优化算法,推荐结果相比于以前基于普通图的推荐算法具有更高准确性.展开更多
The emergence of Event-based Social Network(EBSN) data that contain both social and event information has cleared the way to study the social interactive relationship between the virtual interactions and physical inte...The emergence of Event-based Social Network(EBSN) data that contain both social and event information has cleared the way to study the social interactive relationship between the virtual interactions and physical interactions. In existing studies, it is not really clear which factors affect event similarity between online friends and the influence degree of each factor. In this study, a multi-layer network based on the Plancast service data is constructed. The the user’s events belongingness is shuffled by constructing two null models to detect offline event similarity between online friends. The results indicate that there is a strong correlation between online social proximity and offline event similarity. The micro-scale structures at multi-levels of the Plancast online social network are also maintained by constructing 0 k–3 k null models to study how the micro-scale characteristics of online networks affect the similarity of offline events. It is found that the assortativity pattern is a significant micro-scale characteristic to maintain offline event similarity. Finally, we study how structural diversity of online friends affects the offline event similarity. We find that the subgraph structure of common friends has no positive impact on event similarity while the number of common friends plays a key role, which is different from other studies. In addition, we discuss the randomness of different null models, which can measure the degree of information availability in privacy protection. Our study not only uncovers the factors that affect offline event similarity between friends but also presents a framework for understanding the pattern of human mobility.展开更多
文摘基于事件的社交网(event-based social networks,EBSN)中的个性化推荐服务是一个十分重要且颇具应用价值的问题,现有研究工作主要基于普通图来对EBSN中的关系进行建模,但由于EBSN是一种异构型复杂社交网络,具有多种不同类型实体,因而用普通图建模EBSN会存在高维信息丢失问题,导致推荐质量降低.基于此,首先提出一种基于超图模型的EBSN个性化推荐(hypergraph-based personalized recommendation in EBSN,PRH)算法,其基本思想在于利用超图具有不丢失高维数据信息之特点来更准确地对EBSN中复杂社交关系数据进行高维建模,并利用流形排序正则化计算获取初步推荐结果.其次,又分别从查询向量设置方式改进和对不同类超边施以不同权重等角度,提出了优化的PRH(optimized PRH,oPRH)算法以进一步优化PRH算法所获推荐结果,从而实现精准推荐.扩展实验表明,基于超图的EBSN个性化推荐及其优化算法,推荐结果相比于以前基于普通图的推荐算法具有更高准确性.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.61773091,61603073,61601081,and 61501107)the Natural Science Foundation of Liaoning Province,China(Grant No.201602200)
文摘The emergence of Event-based Social Network(EBSN) data that contain both social and event information has cleared the way to study the social interactive relationship between the virtual interactions and physical interactions. In existing studies, it is not really clear which factors affect event similarity between online friends and the influence degree of each factor. In this study, a multi-layer network based on the Plancast service data is constructed. The the user’s events belongingness is shuffled by constructing two null models to detect offline event similarity between online friends. The results indicate that there is a strong correlation between online social proximity and offline event similarity. The micro-scale structures at multi-levels of the Plancast online social network are also maintained by constructing 0 k–3 k null models to study how the micro-scale characteristics of online networks affect the similarity of offline events. It is found that the assortativity pattern is a significant micro-scale characteristic to maintain offline event similarity. Finally, we study how structural diversity of online friends affects the offline event similarity. We find that the subgraph structure of common friends has no positive impact on event similarity while the number of common friends plays a key role, which is different from other studies. In addition, we discuss the randomness of different null models, which can measure the degree of information availability in privacy protection. Our study not only uncovers the factors that affect offline event similarity between friends but also presents a framework for understanding the pattern of human mobility.