Extracting Precise Activities of Users from HTTP Logs

Browsing histories are often used to build user profiles for browsing supports and personalizations. But, the browsing history also contains HTTP requests generated concomitantly with user activity(concomitant request), which must be removed in order to build correct user profiles. Current filtering methods are based on rather simple characteristics of requests such as the extension of the file name or reported content types. We invent a more efficient filtering method based on other characteristics such as the intervals of requests and the referer relations of requests. In this paper we analyze these characteristics in real web transactions and evaluate their usefulness on filtering.