Detecting individual content-structure patterns in time series data

It is valuable to mine the spatio-temporal patterns from time series data set generated in real world applications. However, two problems of data point granularity determining and user behavior pattern describing are most challenging. To address these problems, in this work, we propose a time sequence based user behavior pattern describing method, in which, the time-related events are mapped on the time stream respectively. Then the user patterns are described according to two aspects, namely the content-based and the structure-based patterns. Based on these two aspects, a new method-CS Similarity, is proposed to measure the similarity of behavior pattern between two independent users'. The experimental results with the real transaction data of Guangzhou smart card show that the proposed method (CSM) has a better performance on finding the similarity pattern among people in compare with the classical methods of WM and SWM.