Abstract: The similarity measure of user behavior based on behavior sequence had not considered the semantic similarity index of activity.To solve this problem,a new algorithm of user behavior similarity measurement was proposed which supported activity semantics.Specifically,the similarity of activities was calculated by combining adjacency relation with label text semantics between activities;the edit weight function and the behavior sequence distance were defined;the users'behavior was modeled with behavior sequence multiple sets and the similarity of user behavior was calculated with Earth Mover's Distance (EMD).The feasibility and effectiveness of the proposed method were verified by comparing with the current mainstream algorithms in terms of measurement properties,satisfiability and experimental evaluation of real data sets.

Key words: user behavior similarity, text semantic similarity, similarity measure, earth mover's distance

摘要： 针对基于活动序列的用户行为相似性度量方法未见考虑活动的语义相似性度量,提出一种支持活动语义度量的用户行为相似性计算方法。首先结合活动间的邻接关系与标签文本语义计算活动间的相似度;其次,定义了活动编辑权值函数和活动序列距离;最后,利用活动序列多重集建模用户行为并利用推土机距离计算用户行为相似度。与目前主流算法在度量性质可满足性、现实数据集实验评估等方面进行对比分析,验证了所提方法的可行性和有效性。

关键词: 用户行为相似度, 文本语义相似度, 相似性度量, EMD距离