摘要:Temporal pattern provides a novel way to character user behavior in social network from the perspective of time. In this work, we study two types of temporal pattern of user behavior in Micro-blog: Long Term Pattern and Daily Pattern. Long Term Pattern stands for the overall trend of user behavior changes since one starts to use Micro-blog and it provides the global view of user behavior variations. Daily Pattern states about the everyday variation and it represents the microscopic regularity within a day. In order to find out temporal pattern, Wavelet Transformation and Dynamic time warping for K-Medoids algorithm (WT-DKM) is proposed to organize time series into clusters, each cluster corresponding to a pattern. Eventually, 4 long term patterns and 5 daily patterns are discovered. These patterns are various in many terms, which reveal the difference of regularity among users. Almost half users behave randomly without apparent regularity while the others’ behaviors have obvious variation trend along with time. User group is often used to character user behavior according to their status, age, etc. So we study the relationship between temporal pattern and user group to discover whether users in the same user group are more likely to behave with same type of temporal pattern. It turns out that for some user groups temporal pattern of most members are identical, while other groups are not.
关键词:temporal pattern;time series;SNS user behavior;clustering algorithms