Research Of Incremental Web Log Mining Based On Som Neural Network And Fuzzy Clustering

With the coming of the era of information explosion, dynamical data updatedmuch faster. Internet users always don’t know how to do when they face vast amountsof data. The phenomenon of so-called "information overload" appears. In order toextract the knowledge people interested from large amount of data, a technology toprovide users with personalized recommendations which can infer people’s requestand preferences based on a series of clicks records what users leave when theybrowse websites gradually become an important issue.Firstly, the characteristics of the classic SOM neural network and fuzzyclustering algorithm were considered in this paper, and a mining algorithm based onSOM neural network and fuzzy clustering algorithm is proposed. By setting a bignumber as the number of output neurons, we can get the rough clustering center setsby using SOM neural network. Then set the rough clustering centers as the inputparameters of the fuzzy clustering algorithm and we can get the clustering centers byusing fuzzy cluster method. Finally the class merging algorithm determines whether acluster need to be merged, and output the final clustering set. Because the improvedalgorithm adopts SOM clustering center as the initial clustering center of the fuzzyclustering, a problem that the multiple random initial clustering centers may appear inthe same cluster which can lead to a bad clustering result is solved. Also because classmerging algorithm is adopted in the algorithm, we do not need to specify the numberof clustering, and the fuzzy clustering algorithm is no longer dependent on thenumber of initial clustering center so much, and different shapes of clusters can befound.Secondly, a web log incremental mining algorithm is designed based on theSOM neural network and fuzzy clustering algorithm in this paper. The algorithm isnot sensitive to noise, and it makes full use of the mining results last time, and is ableto update incremental data quickly. It is suitable for incremental mining the datawhich updates frequently. The algorithm has obvious advantages especially for weblog which is relatively large and frequently updated. Finally, a web log incremental mining model is designed and realized in thispaper. Several experiments were carried out by using the pretreatment of web log datato evaluate the performance of the algorithm proposed in this paper. The experimentalresults shows that the proposed algorithm has a better stability and adaptabilitycompared with the traditional clustering algorithm and it is able to handle theincremental updating problems in dynamic database.