Improvement, Implementation And Application Of Prediction Algorithm In The Recommendation System

Personalized recommendation system has been widely used in the field like network shopping,online music and movie. According to the data of user’s basic information and historical behavior andmachine computation, it can predict user’s interested items and select a certain amount of results for theuser, hoping to win users’ application. In this way, it can urge users to produce consumption behavior.Therefore, the recommendation system has been widely used in the industry. As more and more peopledepend on network in their study and life, it will have practical significance for studying personalizedrecommendation system.At present, the specific application of various recommendation algorithms in different fields hasits advantages and disadvantages. However, the ideas of k-nearest neighbor, similarity and weightedsharing are still the most common thoughts that support and promote the continuous development ofrecommendation system. Yet, with rapid expansion in users and the scale of project information, therecommendation system is faced with technical problems including sparseness, scalability and cold start onthe one hand. On the other hand, it is also faced with practical problems, such as combining with specificfield and utilizing industry information, so as to obtain better results of recommendation.The main content of this paper can be listed as follows: firstly, it studies and summarizes therecommendation system and its main algorithm. Secondly, scoring prediction system without consideringnon-scoring factors will influence on the prediction results. With regard to this problem, the thesis proposesa new scoring prediction algorithm that has considered deviation caused by non-scoring factor. The newalgorithm takes non-scoring factor and scoring factor as two independent terms and evaluates themseparately. It takes scoring prediction as a basic term and non-scoring factor as a correction term so as tocorrect prediction results and improve accuracy of score prediction results. Thirdly, a new algorithm forrecommendation system is proposed, which solves the sparsity problem. The new algorithm turns a sparsematrix into a dense matrix by deleting its isolated points. And the recommendation problem of the isolatedpoints which had been deleted is solved by Association Rule-Based Recommendation.Thus a hybridrecommendation is constructed as a whole, which enriches the diversity of recommendations. And the computational cost was reduced at the same time. Fourthly, the new algorithm will be applied in the filmscoring prediction system. Through simulation experiments and comparing new algorithm and classicalalgorithm’s differences in the RMSE performance, it proves that the new algorithm can improve accuracyof recommendation performance.