Preprocessing Techniques in Web Usage Mining By Using Improved Session Identification Method

Abstract

The phenomenal growth ofthe internethas spawned a varietyof E-commerce applications such as advertisements, enhancement of marketingstrategies and analysis of web usage with regards to modifications and personalizationof website orsystem improvement. In relevance to these applications, mining techniques aidinfindingthe web user’s behavioral patterns. Web usage data cannot be utilized directly as it may lead to inconsistency, incomplete and shortage in certain behavioral patterns and thus it requiresimplementation ofstep-by-step preprocessing techniques. The proposed pre-processing technique involves an efficient User Identification processthatidentifiesthe active users from the global user group. The proposed session identification has been implemented by hybridizing navigation and time heuristic approach to encode the user session browsing behavior in a significant manner. Thus avoid the generation of lengthy unreasonable URLs to improve the quality of the data. The weighted session pageview matrix has been constructed by assigningweights to the pages based on the frequency of the page.Hence, a threshold value is set to remove the lower support session page by achieving the dimensionality reduction to filter the external noise from the data. The proposed session identification method has beenexperimentally tested with a real time data set obtained from the web logfiles taken from the IIS server of Women’s Christian College, Chennai, TamilNadu, India. The experimental results prove that pages having weights less than an empirically determined threshold value have efficiently been removed andthe external noises from the dataset have also been removed successfully.