Fuzzy Temporal Clustering Approach for E-Commerce Websites
G. Sudhamathy, Chinnasami Jothi Venkateswaran · 2012
In this paper a novel approach for clustering of web logs data and to predict intelligent recommendations on the E-Commerce web sites is proposed so as to improve the marketing strategy and to improve customer loyalty. Fuzzy Temporal Clustering Approach (FTCA) performs clustering of the web site visitors and the web site pages based on the frequency of visit and time spent. Time plays a crucial role in the analysis of web usage. Hence these clusters are studied over a period of time to study the migration behaviour of the users and the pages across periods. Such a study can provide intelligent recommendations for the E-Commerce web sites that focus on specific product recommendations and behavioural targeting. Experimental evaluation of the method has proved that this approach FTCA is most efficient, easy to use and a useful clustering approach. Keyword-Web Usage Mining, Web Logs, Clustering, Fuzzy Logic, Temporal Web server records interaction information between the users and the web server in the web log files. This information in the web log files hides user's access patterns and interests and is of great significance for analysis of the user requirements, providing users with personalized services, assisting web personnel and optimizing web sites. Therefore, the web log mining attracts increasingly attention in the fields of science and business. To analyse user's interest on the web pages, clustering techniques are often used in Web log mining. This interest of the users on the web pages is dynamic and they change over a period of time (4). Hence the clusters of users and the pages also change over a period of time. The clustering algorithms can be divided into partition method, hierarchical method, density based method, grid-based method, model based method and etc. The assessments on the clustering algorithm mainly uses two measurement indexes, namely intra class distance and inter class distance. An algorithm that can produce high-quality clustering effect must meet the following two conditions, namely the intra class data or object similarity is the strongest, while inter-class data or object similarity is the weakest. Clustering is a basic understanding activity of human beings. Only through appropriate clustering, the things can be easily researched, and the internal laws of things can be mastered by human beings. The so called clustering is to put things together into a class based on some attributes of things, so that the intra-class similarity is weak as possible and inter-class similarity is big as possible (13).