An approach of splitting web sequential access log for accurate web recommendation

There are many kinds of intelligent technologies for helping web page navigation. Among them, association rules are one of popular technologies to discover interesting and frequent user access patterns in web sites. But association rules have not been very useful in practice because excessive rules are generated. In this paper, we suggest a new approach to discover a small number of more accurate rules from web site access records. By analyzing ten thousands of web page navigation logs from a shopping mall site, we have noticed that people have a special pattern in web page navigation when their interests change. When their interest is changed, they drop by one of frequently visited web pages such as a list page of items or the front page. For example, if a person was visiting web pages on MP3 players and now he wants to move to mobile phones, he usually drop by the front page or a list page of items and then choose a web page on mobile phones rather than directly goes to mobile phones from MP3 players. The proposed method separates a session record into several sub-sessions with such frequently visited pages, and finds user access patterns in the separated ones. The separated sub-session records may have more cohesive access patterns because those are related to the same item. With this idea, we construct a method of web page request prediction. We evaluate our system with huge data set. With those results, we confirm that our proposed method is effective on web page prediction.