Similarity Distance Based Clustering Framework for Aggregation of Web Usage Data

Abstract

The rapid improvement of the World Wide Web as a standard for information broadcasting has engendered a rising attention in the area of Web personalization. Owing to this raise in expansion and difficulty of WWW, web site publishers are facing rising complexity in drawing and maintaining users. To plan trendy and striking websites publishers have got to appreciate their users’ needs. So as to determine the usage patterns that might be analyzed to user’s navigational actions, Web Usage Mining (WUM) application has been used. The knowledge obtained from the study of the user’s navigational actions (usage data) can be expediently subjugated in order to modify the Web information space to the provisions of users. Diverse techniques have been examined in literature for the recognition of Web user profiles. Recent work on web usage mining utilized clustering models such as k-means to aggregate the web user information. K-means is a distance based cluster model becomes more successful if only attribute value ranges are finite and categorical. In case of wide range values and undefined attributes k-means cannot cluster the data efficiently. To enhance the web usage mining process based on users’ behavior, in this work, we generate cluster of web usage data obtained web server logs to aggregate the data access information of the user related to web sites. Multi-cluster model is deployed in our proposal whose distance measure is based on more than one predominant attribute. Similarity distances of dominant attributes are evaluated to its threshold ranges of the web user profile history. Attribute cohesiveness is maintained with web access data logs of past and current data ranges. Web access attributes in consideration for the evaluation are user visits, page visits, session duration, product feature lists, active time of the page access, etc.,. Experimentation is conduct with real data sets extracted from UCI repository. Performance evaluation is made at the three stages to show the effectiveness of the proposed technique compared to the existing one referred from different contributors of web usage mining.