Recommendation Algorithm And Its Application In Social Network

With the development of web2.0, users in social network create a lot of infor-mation. There are too much data so that users can not effectively get the informationthey want. The utilization of information decreases and information overload problemis exacerbated. Currently, search engine can only meet part of user demand, and stillcan not effectively solve this problem. Recommender systems provide very promis-ing approaches to solve this problem. Therefore how to develop effcient, scalable andaccurate recommendation algorithms is a huge challenge.Rating prediction is an important task of recommender systems, in order to buildperfectrecommendationsystems, thisarticlestudiesthealgorithmsinratingpredictiontask from three aspects. First, in order to use other users to do the social recommen-dation, accurate and effcient measure of similarity between users is a key. Based onthe shortcomings of existing methods, we propose counting similarity and similaritytransfer. Through experiments we found that both methods can effectively improve theprediction accuracy and solve scalability problem and data sparseness problem. Sec-ond, based on slope one algorithm, we introduce the idea of machine learning into it.We use stochastic gradient descent method to learn weights and deviations betweenitems, model the user biases and item biases, and make the prediction accuracy greatlyimproved. Finally, we analyzed three time effects in recommender systems : user biasshifting, item bias shifting, and user interest shifting, and then we describe how to in-troduce time effects into two stages of slope one algorithm. By experiments on twodatasets, we found that using time effects carefully can not only enhance the predic-tion accuracy of recommendation algorithms, but also reduce the computational costin some cases.