A New Clustering and Preprocessing for Web Log Mining

Abstract

World Wide Web is a massive repository of web pages and links. It provides information about vast area for the Internet users. There is tremendous growth and development in internet. Users' accesses are documented in web logs. Web usage mining is application of mining techniques in logs. Since due to tremendous usage, the log files are growing at a faster rate and the size is becoming huge. Preprocessing plays a vital role in efficient mining process as Log data is normally noisy and indistinct. Reconstruction of sessions and paths are completed by appending missing pages in preprocessing. Additionally, the transactions which illustrate the behavior of users are constructed exactly in preprocessing by calculating the Reference Lengths of user access by means of byte rate. Using Web clustering several types of objects can be clustered into different groups for various purposes. By using the theory of distribution in Dempster-Shafer's theory, the belief function similarity measure in this algorithm adds to the clustering task the ability to capture the uncertainty among Web user's navigation performance. This paper experiments about the accomplishment of preprocessing and clustering of web log. The experimental result shows the considerable performance of the proposed algorithm.