Research On Collaborative Recommendation Based On User Ratings Data In Social Network

In recent years, e-commerce has accumulated a large amount of data with the development of information technology. Huge amounts of data at the same time for customers to provide diversified products to choose from a selection difficult question. Recommendation system is a personalized push technology based on massive amounts of information retrieval and filtering. It is focused on the actual needs of customers and based on customer's historical data to predict the behavior of user preference and provide personalized information service. However, in the classical recommendation named modeling based on user ratings there are some inevitable problems, such as the accuracy of user similarity calculation, sparse ratings matrix. Aiming at the two common problems this paper has an simple study respectively. In this paper a new user similarity calculation method is proposed and improves the insufficient of the traditional similarity calculation method; According to the item's label a method for unknow ratings predicting is proposed under the condition of rating matrix existing a large number of missing value.In particular, the contribution of this paper mainly includes the following three aspects:(1) For the Pearson similarity in the calculation met the denominator is 0 and similarity to calculate the proposed a improved method. The method can according to user's score to select reasonable similarity calculation method on of is.(2) Based on the existing technology of unknown rating predicting in mahout, combining item label a new method of unknown score predicting is put forward. This method first calculates the user's interest distribution for each label and then predicting unknown ratings according to its labels.(3) The similarity calculation method proposed in view of the new and unknown score prediction method, combining with the mahout the existing methods for better integration.