Research On Recommendation System Based On Collaborative Filtering Algorithm

With the rapid development of the Internet,the amount of information resources that humans have acquired has also increased dramatically.In the face of this information overloaded Internet era,in order to help humans quickly and effectively screen out useful information data,the recommendation system came into being.The core of the recommendation system is the recommendation algorithm.Among the many recommendation algorithms,Collaborative filtering algorithm(CF)is one of the algorithms frequently used by recommendation system.The CF algorithm first collects the user's past behavior data to discover the user's personal preferences,and then groups the users according to the user's preferences,and finally recommends the items that meets their preferences to the target user.However,with the increase in the number of users and the number of items in the e-commerce website,problems such as data sparsity,scalability,computational complexity,inaccurate recommendation results,and poor real-time performance have gradually become the limiting factors for the development of CF algorithm.This article first introduces the development background,working principle of the recommendation system and several recommendation algorithms that are frequently used,describes the CF algorithm in detail,studies the basic idea,workflow,advantages and disadvantages,evaluation standards of CF algorithm,analyzes some of the problems faced by current CF algorithm and the corresponding solutions.Then in terms of user clustering and matrix data filling,this article makes the following improvements to the traditional CF algorithm:(1)Aiming at the problem that the traditional K-means algorithm randomly selects the initial clustering center and the initial k-value,which is easy to lead to inaccurate recommendation results.Based on the existing improvement work of the initial clustering center,this article proposes an improved method of selecting the initial clustering center and k-value.This method is based on the idea of the local optimal solution of the minimum spanning tree,uses the improved Kruskal algorithm to cluster the users and generate K user clusters to obtain more accurate user clustering,the algorithm can effectively alleviate the scalability problem and reduce the amount of computation.(2)Considering the problem of sparse data,this article proposes an improved data pre-filling algorithm.Based on the improved Tanimoto coefficient(generalized Jaccard coefficient),the algorithm obtains a more reasonable user similarity and proposes a new prediction formula,calculates the score value of the target user for the unevaluated item,and adds the value to the user similarity matrix,which reduces the sparseness of the matrix data.(3)Adopt a cascading combination to organically combine the improved K-means user clustering algorithm with the improved data prediction filling algorithm.The combination algorithm solves the problem of data sparsity and scalability and user cold start,reduces time complexity and improves the accuracy of the recommendation.Finally,this article uses Movie Lens dataset as experimental data to verify the effectiveness and enforceability of the improved collaborative filtering algorithm proposed in this article.