A novel competitive neural network for Web mining

A novel neural network named competition-based self-stimulative network (CSN) is proposed in this paper. Firstly, each input vector is presented to the network to compare with the prototype vector that it most closely matched. If there are not any prototype vectors to match the input pattern, a new prototype is generated to hold it. Then Hebb rule is used to activate neurons simultaneously in the second scanning. If more than two prototype vectors match, the strength of the synapse will increase. After activated neurons in competitive layer stimulate each other, neuron fuzzy graph made up of neurons of competitive layer is formed. Finally, making λ cut graph to get connected components, neurons in the same connected components of λ cut graph export to the same neuron of output layer created currently. Being not sensitive to the input sequence of data and the noises ART existing, it can either hold the former memory, or memorize the new patterns. Web log files mined by this algorithm can dispose noises and get good results of the clusters in users and pages. Not only can it help us to discover the user access patterns effectively, but it can provide the valid decision-making for the Web master to devise the personalized Web site. Our experiments on a large real data set show that the approach is efficient and practical for clustering of Web users and pages.