Abstract
The mobile Internet is characterized by “Easy-come and easy-go” characteristics, which causes challenges for many content providers. The 24-hour clickstream provides a rich opportunity to understand user's behaviors. It also raises the challenge of having to cope with a large amount of log data. The author proposes a stream-mining oriented algorithm for user regularity classification. In the case study section, the author shows the case studies in commercial mobile web sites and presents that the recall rate of the following month revisit prediction reaches 80-90%. The restriction of the stream mining gives a small gap to the recall rates in literature, but the proposed method has the advantage of small working memory to perform the given task of identifying the high revisit ratio users.
