基于签到数据的用户空间出行相似性度量方法研究

Clustering LBSNS Users Based on Check-in Records

近年来,国内外类似街旁、人人、Foursquare、Gowalla等基于地理位置的移动社交网络(LBSN)发展迅猛,大量用户通过这些服务以签到的方式记录时空行为轨迹,这些个体行为轨迹数据为我们研究用户行为模式以及探究其内在规律提供了巨大的机会和挑战。然而,LBSN用户的相似性并没有从地理位置以及用户轨迹加以考虑,本文提出了基于格网划分的方式对用户空间出行进行相似性分析,通过用户轨迹建模以及相似序列匹配,探索用户出行轨迹的空间相似性度量方法并评估相似权重,最后通过用户好友关系与相似性权重的比对,证明了该方法的有效性。

LBSNS（Location Based Social Network Service） like Jiepang, Qieke, Renren Places, Foursquare and Gowalla support hundreds of millions of user to log their footprints through check-ins with spatio-temporal data. The increasing availability of large amounts of global-scale footprints data pertaining to an individual＇s trajectories bring us opportunities and challenges to model patterns of human mobility and automatically discover valuable knowledge from these footprints trajectories. In this paper, we aim to geographically mine the user similarity and explore particular user classification based on their footprints. Even though human movement and mobility pattern have a high degree of freedom and variation, they also exhibit strnctural patterns due to geographic and social constraints. Hence, we devise a framework to model each individual＇ s footprints, measure the similarity among users and explore the particular user classification. In this framework, we took into account the sequence property of user＇ s movement and evaluate this framework using the check-ins data of 602,239 users and 8,383,949 check-ins. Such user similarity is significant to individuals, communities and businesses by helping them retrieve information with high relevance.