Evaluation of web usage mining approaches for user's next request prediction

Abstract

Analysis of Web server logs is one of the important challenge to provide Web intelligent services.In this paper, we describe a framework for a recommender system that predicts the user's next requests based on their behaviour discovered from Web Logs data. We compare results from three usage mining approaches: association rules, sequential rules and generalised sequential rules. We use two selection rules criteria: highest confidence and lastsubsequence. Experiments are performed on three collections of real usage data: one from an Intranet Web site and two from an Internet Web site.

A Layered Approach to Revisitation Prediction

“…A session S is the bag of all pages p i visited by a user u in the same browser tab for a time period of up to 25.5 minutes ( [8,17]), placed in chronological order, from the earliest to the latest one: S = {p 1 , p 2 , . .…”

“…Sequence mining techniques constitute a variation of this approach, in the sense that they do not consider the strict order between items [2,15]. A comparison of these techniques with AR was conducted by Géry and Haddad [8], with the outcomes of their evaluation suggesting that Frequent Sequence Mining has the best performance. Nevertheless, all these methods still suffer from the inability to predict/recommend unseen items (i.e., not included in the training set, typically due to concept drift).…”

Mining Indirect Association Rules for Web Recommendation

“…In the typical, item-to-item approach to recommendation based on association rules, ranking lists are created from the entire set of direct rules d i →d j that exceed minimum confidence and minimum support level (Chun et al, 2005;Géry and Haddad, 2003). The pages d j from all rules d i →d j outgoing from d i are considered at the creation of recommendation ranking lists for the page d i .…”

“…In another approach to recommendation, the system retains user profiles or any other kind of historical or recent information related directly to the particular user. Based on these data, we can personalize recommendations according to either past (Adomavicius and Tuzhilin, 2001;Cho et al, 2002) or present user activities Géry and Haddad, 2003). Nevertheless, in this paper we focus only on non-user-sensitive recommendations, which enable us to create a static list of preferred pages individually for each web page.…”

“…Association rules were utilized in many web recommendation systems, applied in various domains such as suggestions in personalized distance learning (Wang et al, 2002), or next steps in web navigation, i.e., hyperlink recommendation (Géry and Haddad, 2003;Mobasher et al, 2000;Yang and Parthasarathy, 2003), personalized shopping adviser (Cho et al, 2002;Chun et al, 2005;Ha, 2002), the extension of web searching (Boley et al, 1999). In the web environment, association patterns can be extracted from server logs (Géry and Haddad, 2003;Mobasher et al, 2000;Wang et al, 2002;Yang and Parthasarathy, 2003), purchased products (Cho et al, 2002;Chun et al, 2005;Ha, 2002) or products placed into the basket (Cho et al, 2002), as well as the web content (Boley et al, 1999).…”