Research On User Access Pattern Based On Web Log Mining

One of the most important fields in database is data mining.in view of its wide application and pracrical significance,the technique and application of data mining developed rapidly and attracted much more attenting both in fields of academic research and information industry.Discovering the interested,hidden and unknown data from large data sets is the purpose of data mining. The main work of data mining is to deal with the structural data,while the web data mining is based on Internet to get the interesting and potential pattern from the half strutual or not structral web pages.Data in Internet is a half structural system,and it is difficult to dial with them.Fortunately,the web sever log files have a nice structure and it is very convenient for data mining.Furthermore,web log mining is a branch of web usage minig and has special theory and practice significance as an important part of web mining.In this thesis,the precess of data mining,web data mining and web log mining was repoted,Focus on the web log minig,the method and technology of web log mining were discussed in this thesis.Analysed and expounded web log minig especially discussed prediction method of Markov model.The exsisting metheds have some disadvantages such as the precision is not hige enough.We promote a new prediction method which also use Markov model named prediction method of Markov model with access trend weight. We discovered that those prediction methods just consider the counts of browsing page,resulting in several results,we can say that the precision is not high enough.If the time of browsing page is taken in consideration,the problem can be resolved to some extend.The new prediction method has the following features:the parameter of broweing time is put in Markov model,and they were discreted by using Value Difference Metric method,proposed a new definition Acceess trend weight,and use it to promote the caculating method of transaition matrix,a new prediction algorithm is also been proposed, then the mothod is reslized by experiment.It has been proved that prediction method of Markov model with access trend weight has higher prediction precision than the previous one.At last, we come up with a logic frame of a Web access model mining system.