Recommendation System Algorithm Based On Distributed Random Optimization Method

As the widespread of social network media like Facebook, Weibo and so on, our daily life are filled with variety kinds of information, the friends in social network and large amount of knowledge when we surfing the Internet. How to find the most useful and valuable information becomes a more and more import question for the social network researchers, when it comes to the large amount of user features and the log record the traditional methods may turn useless and non-effective. Hence the blending of the algorithms in recommendation system came into use. In this thesis, the idea of collaborative filtering to select features in social network relationship is proposed, which select the features to build a vector space model. Then a distributed algorithm based on factorization machines and stochastic gradient descent algorithm is proposed, using the sparse matrix as one part of input. When it comes to the filter of train dataset, the records are split into different sessions depending on the time sequence. All the methods mentioned above have proved that have remarkable improved the recommendation accuracy.In the part of experiment, the Weibo of Tecent and Sina are used for data analysis and verification, The result shows the availability of our algorithm and also proved the useful of the combination of distributed computing and recommendation system.