||Data mining techniques, mainly clustering and association rule mining, are widely applied to the World Wide Web scenario. Various applications including server log mining, collaborative personalization are proposed and implemented. In majority, these web mining approaches are concentrated on helping a specific content provider or a website in restructuring or in tailoring its contents for each individual reader. This thesis proposes another approach to provide assistance to end-users instead. A detailed client side user activity log is captured every time the user browses the web. From this activity log, the user’s interests are discovered and categorized using clustering techniques; the user’s web browsing habits are discovered and formalized using association rule mining. Such knowledge forms personalized knowledge base. Using this knowledge base, an intelligent agent, called WBext, is able to help the user to search the web by extending a simple keyword query. By observing and comparing user activities with existing knowledge about user’s habits and interests, the agent is also capable of recommending links to the user while he/she is browsing the web. The effectiveness of the approach is evaluated based on a series of metrics; results of various experiments are presented and analyzed.