Efficient mining of temporal traversal patterns from very large Web logs

A Web page in a Web access session is considered as a temporal content page, if its access time is greater than the average access time of all the pages in the session. A maximal temporal reference of a Web user in an access session is a longest consecutive sequence of Web pages in the session which ends at a temporal content page and has no other temporal content pages in the sequence. The problem of efficient mining of frequent temporal traversal patterns, i.e., large temporal reference sequences of maximal temporal references, from very large Web logs is important in Web mining. This paper aims for algorithmic solutions to the problem with best possible efficiency. We first design linear time algorithms for finding maximal temporal references from Web logs. We then devise a linear time algorithm for mining frequent temporal traversal patterns, utilizing the technique developed in [8, 9] for fast construction of "shallow" generalized suffix trees over a very large alphabet.