Mining and tracking evolving Web user trends from very large Web server logs.

Online organizations are always in search for innovative marketing strategies to better satisfy their current website users and lure new ones. Thus, recently, many organizations have started to retain all transactions taking place on their website, and tried to utilize this information to better understand and satisfy their users. However, due to the huge amount of transaction data, traditional methods are neither possible nor cost-effective. Hence, the use of effective and automated methods to handle these transactions became imperative. Web Usage Mining is the process of applying data mining techniques on web log data (transactions) to extract the most interesting usage patterns. The usage patterns are stored as profiles (a set of URLs) that can be used in higher-level applications, e.g. a recommendation system, to meet the company's business goals. A lot of research has been conducted on Web Usage Mining, however, little has been done to handle the dynamic nature of web content, the spontaneous changing behavior of users, and the need for scalability in the face of large amounts of data. This thesis proposes a framework that helps capture the changing nature of user behavior on a website. The framework is designed to be applied periodically on incoming web transactions, with new usage data that is similar to older profiles used to update these old profiles, and distinct transactions subjected to a new pattern discovery process. The result of this framework is a set of evolving profiles that represent the usage behavior at any given period of time. These profiles can later be used in higher-level applications, for instance to predict the evolving user's interest as part of an intelligent web personalization framework.