Research On Web Usage Mining Based Users Consuming Patterns Discovery

Web technologies have found many and wide applications like information sharing, E-Business and online service. Many companies invest a great deal of money in constructing their Websites to issue their messages, making Ads for their products and providing services on other's Websites or doing E-Business on the Internet. These companies are wondering their fund utilization rate urgently in order to improve their business strategies to catch more business chances and provide better services to their users. Hence, it is very important for companies to acquire and understand the consuming behaviors of their users.Having made analysis and researches on web usage mining systemically based on Web log records and according to some conventional Web mining models, four new data mining models are issued and applied to Web usage mining. A web usage mining system is designed and implemented to find a user's personal consuming patterns.The works in this paper are arranged as follows:1. Feasibility analysis is made and difficulties of research on Web usage mining are pointed out. The process of data pretreatment is introduced and a new algorithm called 'father nodes finding for complement' is issued for path complement.2. A new model considering user browsing process adequately is issued to construct similarity matrix of web pages which shows similarity degree for any two pages in a website and makes pages clustering more rationally.3. Conventional Web mining models for user most frequent navigation patterns almost only consider the distance between pages and ignore the structure of the Web site, so they can't do data mining accurately. A new web mining model for user frequent navigation patterns which considers the Web site framework adequately is issued. It is proved that the model figures out the shortages of conventional Web mining models for user frequent navigation patterns by experimentations.4. For users clustering, a new model is issued to construct user transition matrix for Markov based on conventional model, and it considers the Web site structure adequately and makes Web usage mining more accurately. The results of users clustering and user most frequent navigation patterns are integrated successfully to find interests and taste of a user group, and then it provides ample warranty for companies when they make business strategy.Finally, a visual Web usage mining system is implemented with JBuilder. And the four new models are applied in this practical system. The test results show that the new models are very feasible and effective by experimenting based on the Web log records in Websitefhttp://202.118.69.137:8000) in our Lab.