K-means Cluster Algorithm Based On Improved PSO And Its Application In Recommendation System

With the era of big data coming,the information accessible to human beings is becoming larger and larger,but much of the information may not be what we want.In order to solve this problem,clustering analysis and recommendation system emerged at this moment.Cluster analysis divides the mass of information into categories according to a certain feature,so that we can search for the information of interest in the categories we need.K-means clustering has the advantages of simple principle,easy implementation and fast convergence,and is usually used to deal with some clustering problems.However,the use of the algorithm is limited to the disadvantages of the algorithm,such as the vulnerability to the initial center point and the inability to determine the k value in advance.Particle swarm optimization(PSO)has the advantages of fast searching speed,easy implementation and insusceptible to the effects of the initial center point.It is often used to solve optimization problems and combine other algorithms to achieve more efficient.The idea of the collaborative filtering recommendation algorithm analyzed the preferences of users based on the information of past actions of users and make personalized recommendation.the realization of the algorithm is simple and efficient,but it has the inevitable problem of data sparseness and cold start.Moreover,the scalability of the algorithm is poor.First,the paper studied the K means clustering algorithm and particle swarm optimization algorithm.Aiming at the problem of the premature convergence of particle swarm optimization algorithm and inability to dynamically adjust the weighting factor,the particle swarm algorithm is improved by proposing the chaos search process and improved ideas of adaptive adjustment of factor.The paper puts forward the K-means clustering based on improved particle swarm algorithm(IPK-means)further,which searches the better initial center for K-means by using the improved particle swarm optimization algorithm which is insusceptible to the initial center.Experimental results show that K-means algorithm based on improved particle swarm optimization has better clustering effect.Then,the paper studied the collaborative filtering recommendation algorithm and proposed following improvements according to its disadvantages :(1)the user-attribute matrix is used to replace the user-movie matrix to reduce the data sparsity problem;(2)adding the IPK-means algorithm to divide the cluster of users to increase the scalability of the algorithm;(3)adding ebbinghaus curve to simulate the user's interest changing with time.The results of experiment show that the improved algorithm not only has more accurate recommendation results,but also more stable recommendation results.