Research On Web Log Mining Based On Concept Lattice

With constant increment of Web resource, the framework of Web sites become more and more complicate nowadays. The present Web sites are not intelligent and active, so it's very difficult for users to find their needed information rapidly in front of such huge and complex World Wide Web. Meanwhile the design of Web sites becomes very hard. So Web sites should be improved to provide better and intelligent Web services, such as Personalized Web Service, Adaptive Web Site and so on. One of the most important reliance of improving Web sites is user access information, which is recorded in Web log files format, so Web log mining can discover user access patterns efficiently and therefore resolve the problems above.In this thesis, the process of Web log mining including data preprocessing, patterns discovery and patterns analysis is discussed. We analyze all tasks of data preprocessing and kinds of pattern discovery technology. We analyze the deficiencies of Godin algorithm that it needs to traverse all the concept nodes to find the generator of the new concept, therefore propose a new algorithm based on tree structure to update concept lattice, which can reduce the search-range of generators and parent nodes, so it has better efficiency than Godin algorithm. We apply concept lattice theory to discover patterns in Web log mining, which can resolve incremental mining problem. We research association rules mining based on concept lattice, optimizing the generation of strong association rules by maximun frequent itemsets, which can reduce the time of finding frequent itemsets and unnecessary rules, therefore improves the efficiency of patterns discovery.The advantage of incremental concept formation algorithm is that it can update the original concept lattice so that our method of pattern discovery is meaningful to update the pattern database of intelligent Web services.