Improvement Of Collaborative Filtering Recommendation Algorithm Based On User

With the popularization and development of the network, the recommendation system is more and more used in the practical application of the business.More and more enterprises have seen the value of the recommended system. However, in the application process, there are some problems such as data sparsity, scalability, user interest, and the transfer of time or space. In view of problems of recommendation system in the practical application, this paper analyzes the collaborative filtering algorithm based on users, found separately improved methods from the user and project dimensions of this algorithm. In user dimension, this paper proposes a collaborative filtering recommendation algorithm based on user's local nearest neighbor and category preference, which can solve the problem of user interest transfer and data sparsity. The algorithm using K-mean clustering algorithm is the first project for the classification and again in each category were analyzed, the calculation in each category similarity of users, thus obtains the user in each category of recommended items. Three calculation methods are proposed to calculate the degree of preference of the user, and the user's preference of each category is taken as the weight. Because the algorithm considers the problem of local similarity and the category preference of users, the improved algorithm has better recommendation effect. In the project dimension, this paper proposes a collaborative filtering recommendation algorithm based on nearest neighbor association rule mining algorithm to solve the problem of data sparsity. The algorithm firstly finds the nearest neighbor of the target by the similarity computation, and then carries on the association rule mining to the neighbor user's score project, finds out the correlation between the items. In the process of mining association rules based on FP Tree, fixed frequent itemsets number, dynamically adjusts the minimum support degree, and degree of ease based on user collaborative filtering recommendation algorithm that in dilute data meet the recommendation quality. Generate a set of recommended items according to the existing scoring items and the associated rules. Then according to the similarity of the user to recommend items in the collection of items in the forecast, the higher the score of the first N project recommended to the user. The experimental results show that compared with the traditional recommendation algorithm, collaborative filtering recommendation algorithm based on nearest neighbor association rule mining has a certain improvement to the accuracy of the recommendation system, which improves the quality of recommendation.