The Research Of Recommendation Algorithm Based On Modified Equal Probability Spreading

With the increase of web data, people can search many kinds of informationthrough network. However, in face of such large and complex information, it is difficultto easily find what they are interested in. This makes the recommendation system moreand more popular. Through the recommendation of recommender system, people canfind the information in accordance with their preference. Currently, the recommendationsystem has been applied into the e-commerce, social media, film and other fields.In this paper, first of all, we introduce the traditional collaborative filteringalgorithm (CF), recommendation algorithms which are based on back propagationneural network (BP) and support vector machine (SVM). In the CF, we mainlyintroduce the several calculation methods of similarity and prediction score. In the BPand SVM, we mainly introduce how to build recommendation models of the BP andSVM. By observing mean absolute error between prediction score obtained by eachalgorithm and actual score, we assess the performance of each algorithm.Then we study two modified algorithms of Equal Probability Spreading algorithm(EPS) on the basis of bipartite graph, called Adjusting Resource Allocation EqualProbability Spreading algorithm (ARA-EPS) and Redundancy-Eliminated EqualProbability Spreading algorithm (RE-EPS). In the ARA-EPS, we optimize products’resource allocation of the EPS algorithm. Research result shows that, comparing withthe EPS algorithm, the accuracy of the ARA-EPS algorithm is slightly increased. Thenwe further study the effect of redundancy attribute on the algorithm performance, andpropose the RE-EPS algorithm. Research result shows that, comparing with the EPSalgorithm, the accuracy of the RE-EPS algorithm is increased by21.2%, and thediversity also gets a certain degree of increase. That is, the RE-EPS algorithm furtherimproves the performance of the EPS algorithm. In addition, the RE-EPS algorithm ismore advantageous than the CF and ARA-EPS algorithm in recommendation accuracyand diversity.