Optimization For Collaborative Filtering Recommendation Algorithm And Test System Implementation

The development of the Internet allows users to drown in the vast information, and users want more personalized service, therefore, the recommendation system begin to develop. Recommendation system can be achieving targeted recommendations based on the user's browser, ratings and other data to enhance the user experience, while improving user stickiness of website. Recommendation algorithm is the core of the recommendation system, and directly affect the recommendation system's accuracy and stability.Therefore, to enhance the performance of the algorithm can improve the accuracy and stability of the recommendation system.Collaborative filtering algorithm can be an important branch of the recommendation algorithm because of its simple, efficient and useful, therefore, Slope one has been seriously as a classic collaborative filtering algorithm and developed the IS-Slope one based on the project similarity and the SVD-Slope one based on singular value decomposition.However, there are still existing cold start and sparse data problems during the recommendation process and seriously affecting the accuracy and stability of the recommendation system.This paper's demand points are to improve the accuracy and stability of the recommendation system. The author analyses the main problems and proposes practical solutions to every problem;Then through the study of algorithms and programs, proposing collaborative filtering algorithm US-Slope one based on user similarity;In the algorithm design process, the user similarity calculation algorithm had been studied and preferred, and the sparse matrix storage structure had been studied and preferred;Finally,the article use the software engineering as the guide, designing a user-oriented similarity recommendation system, and demonstrating the performance of the algorithm which had improved from horizontal and vertical angles respectively.This paper selected high reliability and high validity data as the sample space, and also chose the mean absolute deviation as test guidelines to test algorithms.In the horizontal test, the Slope one, IS-Slope one, SVD-Slope one, US-Slope one were tested on MovieLens data sets,verifying the US-Slope one recommendation algorithm's performance boost;In the vertical test, different sparsity data sets were tested,verifying the US-Slope one algorithm's recommendation stability under different sparsity circumstances.This paper mainly focuses on collaborative filtering algorithm --Slope one optimization and designed recommendation system,proving the degree of optimization of performance.First of all, by studying recommendation algorithm and it's main issues, summed up the recommendation algorithm's modeling thought and it's advantages and disadvantages,then put forward the user-oriented similarity collaborative filtering algorithm --US-Slope one recommendation algorithm.For the performance test, it is necessary to use a high degree of reliability and validity data set.In order to test the algorithm recommended thoroughly, the paper achieve performance testing from both horizontal and vertical angles to ensure the accuracy of the recommendation algorithm performance.