A Kohonen Neural Network Based Model For User Access Pattern Mining

Kohonen neural network (KNN) is a self-organizing feature map withoutsupervision. KNN features training weights through competitive learning, andobtaining center points of clusters in an automatic manner. Due to these twocharacteristics, it is widely used in pattern recognition, pattern control, and etc. By thesame token, it is employed in the user access pattern mining.User access pattern represents users' interests in websites. Relying on analysisand discovery of the pattern through user log files, it can improve the performance ofthe web server, reconstruct the website, find the potential customers of e-commerceand enhance the quality of web services.However, KNN is not good enough when directly used in user access patternmining. KNN produces only the best matching node, whose Euclidean distance fromeach weight vector to the chosen sample vector is minimal among all output nodes.That is to say, only one interest will be found if it is applied in user access pattern.Obviously this result many other users' interests being neglected. A possible solutionaround this problem is to use triangle membership function instead of choosing onlythe best matching node.With the improved algorithm, a user access pattern mining model was designedand implemented. One difficult is: how can the log files be converted to samplevectors as they are two different data format? This problem can be addressed byintroducing a database as buffer.Finally, some experiments are taken to check the effectiveness of the proposedmodel and comparisons with other models are made. The experiments show that ourmodel is effective in mining user access pattern.