Mining Indirect Associations in Web Data

ABSTRACT Analysis of asso iation is an important Web mining te hnique be ause it an provide useful insight into the navigational behavior of Web users. E-tailers an use this information to develop strategi marketing plans and to re-stru ture their Web site in order to enhan e the browsing experien e of their ustomers . Previous work on mining Web asso iations has fo used primarily on nding frequent a ess patterns in the data. These patterns an be generated by Web users who share similar information goals or by those with varying interests. Sin e Web asso iation patterns onsider only o-o urren es in data, it is diÆ ult to identify patterns generated by one group of Web users but not by the others. Another drawba k of the existing approa h is that it does not adequately address the impa t of Web site stru ture on the support of a Web page. As a result, the majority of Web asso iation patterns dis overed using onventional te hniques ontain the home page or other referen e pages that have multiple outgoing links. In this study, we apply a new mining te hnique alled indire t asso iation to Web usage data. This novel te hnique is apable of ombining the various asso iation patterns into a more ompa t stru ture. It an also apture both positive and negative orrelations that exist in the data. We demonstrate the appli ability of this te hnique on Web data from both ommer ial and resear h institutions. Our analysis shows very promising results, espe ially in terms of identifying Web users with distin t interests.