The Application Research Of Collaborative Filtering On Recommendation System

We are living in the era of information explosion, the data produced by humans and machines increase by terabytes every day,the human society gradually enter data age from IT. Information overload and interested data mining has become two big problems human face, in order to effectively solve the two problems, academia and industry puts forward the concept of recommendation system.Recommendation system is mainly composed of three parts:data acquisition, the recommendation engine which is the core part of the system and data visualization. In the face of mass user behavior data, the traditional recommendation systems and algorithm has great limitations on real time, the quality of recommendation and so on. In order to solve these problems, universities and major Internet company have introduced the manpower and material resources to carry out research, and achieved some good results.The main work and innovation of this paper include four aspects:Firstly, this paper studied the classical collaborative filtering algorithm including item-based、user-based、SVD、Graph model and Clustering etc, and focus on the core principle, operating mechanism advantages and disadvantages of the algorithm.Second, this paper use the domestic mainstream business platform Dangdang and Jingdong Mall as the application case of recommender systems, focuses on the analysis of the recommendation system of two platforms is how to carry out the recommendation, and improve service experience.Third, through the research of the core principle of the classical recommendation algorithm, the recommendation angle and the advantages and disadvantages、it put forward a hybrid strategy recommendation algorithm based on clustering model,the algorithm combines online and offline calculation,from project to the user,use the similarity between goal user and user which had scored on the project as the weight, to calculate the user score on the project, then add item rating data to the user’s score list,finally, based on the user rating list, it use the TopN algorithm to recommend projects to users. The algorithm is improved in the aspects of scalability, data sparsity and real-time recommendation.Fourth, the analysis and verification of improved algorithm, this paper selects domestic and internationally recognized the data set MovieLens for experiment, use the MAE as the evaluation standard of the algorithm, analyzes the recommendation quality of the algorithm in different nearest neighbor number k and the training set/data set ratio R value.In addition, based on improved hybrid strategy recommendation algorithm it developed the Web Recommendation System which its data use the MovieLens, and through the precision, recall, and F1 category indicators, it further validates the quality of improved algorithm.