Personalized Message And User Recommendation In Social Media Based On Statistical Learning

In recent years, as the Internet to flourish, especially accompanied by therise of social networks websites, it was discovered that the Internet began toappear on the phenomenon of information overload, too much information willnot help people more easily to find information, but more difficult, from the largeamounts of information, people cannot find out what is important and what isoptional, and social networking is a self-media network applications, anyone, atany time can post a tweet on the website, apparently to increase the degree ofinformation overload.Based on statistical learning personalized Recommendation in Social Mediause statistical machine learning method to establish personalized user model, tohelp users avoid information overload, the model discovery important tweet andfriend for microblog user, this technology is also very important for socialnetwork websites to improve user experience, at the same time, the recommendersystem and social networking is the hot research field currently, we believe thatthe recommendation in the social networking media is also important.The main content is divided into tweet recommendation and friendrecommendation two parts. For tweets recommendation, we propose apersonalized method to push the text content to the user, we give the featureanalysis and experiments show the effectiveness of our method. For friendsrecommendation, we hope that by friends recommended carefully for users ofsocial media, they will be able to choose the right friends and thus able toachieve the effect of information filter. We proposed several methods based oncollaborative filtering, heuristic, link prediction, topic model. Finally, in order toboost the prediction accuracy of friends recommendation, and to take advantageof multi-models, we combine all of models by using ensemble learningtechnology.The experiments proved the effectiveness of our proposed methods, for theinformation recommended, we get the recall rate0.49, for the userrecommendation, we examined the performance of a variety of recommendationalgorithms, and boost performance through the ensemble learning.