Research Of Item Clustering Based On Particle Swarm Optimization In Recommender Algorithm

With the rapid development of Internet, a large amount of information has emerged on the Web. Information overload problem and information lost problem is becoming increasingly seriously. E-commerce technology provides users with many choices. Web structure has become more and more complex. Users frequently lost in many goods and can not successfully find goods they needed. In this case, Web data mining and e-commerce system become combination to be a branch of data mining - E-commerce recommender system. E-commerce recommender systems directly interact with the user. They simulate sales staff to recommend goods to customers and help users to find goods they really need and the web sites can maintain web users, improve sales and their service quality.As the result of good prospects, E-commerce recommender system in theory and practice has been greatly developed. But with the continuous expansion of business scale and the rapid growth of information and goods categories in the Internet, recommender systems have to face the challenges. As the challenges of e-commerce systems, the paper did the research on recommender system and recommender algorithm, then a new item clustering algorithm based on particle swarm optimization algorithm was proposed in recommender system.Particle Swarm Optimization algorithm is an evolution of computing technology. It has the characteristics that are simple, effective, fast convergence, great global search ability. In recent years it was great concerned by academic. In the past, item clustering algorithm of recommender system was based on the k-means clustering algorithms, but the shortcomings of it are easy to fall into local optimum, efficiency is not high, and measuring the similarity was not accurate in the recommender system. Therefore in this paper, particle swarm was introduced to optimize clustering process. The fitness function of particle swarm optimization can be more accurate to measure the similarity between items and quickly find better cluster center. In the study, as recommender algorithm has the questions that the data is sparse, the means value is used to lower the sparsity. Then clustering algorithm based on particle swarm optimization was used to generate the cluster centers. After this the nearest neighbor was searched through cluster centers and a recommender was generated. The algorithm improved the accuracy and the real-time performance, lower the sparsity. Experiment results indicate that the algorithm can effectively improve the accuracy of the recommender system.