Research Of Algorithm Of User Patterns Mining Based On Self-organization Neural Network Of Fuzzy Clustering

At present, the rise of personalized information recommendation technology has become one of the most popular areas of current Web usage mining. Personalized information recommendation is based on Web users to browse the site’s Web logs and other user information, suggesting that their preferences and interests, and provide personalized information services. This paper studies personalized recommendation in the process of user clustering algorithm, in order to gain user interest in a variety of Web log analysis.This paper represents a merging clustering algorithm. The algorithm based on the set of clustering center can determine the number of clusters of sample set. First, Implemented by a self-organizing neural network algorithm, use the default of a large number of clusters to learn and train the network to get the set of cluster centers. Merge them on the set of cluster centers by this algorithm, using the cosine formula to calculate the similarity of the cluster center, build the evaluation function to judge the cluster center to meet the combined requirements, and finally attached to the samples of the cluster center. The degree of distribution is uniform to combine the set of cluster centers, resulting in a sample set of the new cluster center sets and the number of clusters.This paper, fuzzy C-means clustering algorithm is applied to the self-organizing neural network topology; a fuzzy clustering neural network is created. Use of self-organizing neural network topology and training ability to learn, to effectively address some of the large amount of data operations in the fuzzy C-means clustering is slow and center the weight initialization problem. At the same time, the fuzzy C-means clustering in dealing with complex on the question of its increase in the characteristics of fuzzy clustering, a marked improvement in the clustering effect, and can be tapped to the user’s interest.Stage in the application of fuzzy clustering neural network, the competitive fuzzy layer neurons weights set to the weights before the training phase, and remained stable, no longer be updated to adjust. According to membership functions, update each input sample to the membership of the output neurons. Learning rate is adapted adjustment according to the degree of membership, can avoid selection and adjustment of the neighborhood. Finally, according to the distribution characteristics of the membership degree, setting the output threshold, the input samples to the membership of a given output neuron is greater than the threshold, the output of the neuron corresponding to the category information.Finally, the mining models by using the design of training on the pretreatment data, the model can automatically clustering. Then randomly drawn from the input sample data to test and evaluate the performance of the model experiments show that significantly improved fuzzy clustering neural network clustering effect, accurately depicts the fuzzy characteristics of the user, the user’s multiple interests and preferences.