Research And Application Of Hybrid Recommendation Algorithm Based On SNS

With the rapid development and extensive application of social networks, people can get a flood of information through the Internet; as a result, information overload has become one big issue need to be addressed. Recommendation technology is an important method to solve the problem of information overload; in the meanwhile, this technology has being an important research subject in the domain of natural language processing and data mining. Obviously, recommendation technology has high academic and applied value in many fields. In this thesis, a hybrid recommendation method based on content and collaborative filtering is used to complete the recommendation of the related users and related information. The main work of this thesis is as follows:1) This thesis first presents the research of similar words and related words based on language model. Combined with the analyses of Chinese part of speech, it constructs the analysis model of similar words and related words by utilizing users’ historical data, and thereby gets the POS-CBOW language model and POS-Skip-gram language model.2) Secondly, this thesis proposes the words and persons recommendation based on language model. By analyzing the linguistic of text and the high dimensional spatial word vector of the language models, it completes the calculation of related words and persons.3) At last, this thesis presents the recommendation of user community and related information based on the hybrid recommendation. Through the language model with shallow semantic relations, our method can get the semantic stratified words first, and then get the semantic stratified users by using the semantic stratified words. Finally through the emotional computing and interest word extraction, this thesis can complete the user’s collaborative filtering recommendation.The experimental results validate the methods in this thesis has the value of research and application, such as the recommendation application of Sina Weibo, and the relevant information recommendation of the full text retrieval. In addition, our work also has some insufficiencies, and the Cold-start problem in this thesis is still one of the important tasks in the future.