Research Of Personalized Recommendation System Based An Mixed Model

Nowadays, the Internet is the main source to get information for people, and all aspects of people's lives have been connected with it, which makes people's life more convenient, but also brings a series of problems at the same time. We usually lost ourselves in the Internet because of a lot of heavy and complicated miscellaneous information. We can't find what we need or take a long time to find what we need in the internet. It is very inefficient by using the traditional search engines, so it is difficult to meet the users' needs, naturally, Recommender System was born. Recommend system can provide personalized services from person to person and enhance the user's trust to the sites.Music recommendation system provide the most suitable music to the user according to user's interests and hobbies, accordingly this can enhance the user of the site's trust.It is not difficult to find that there are weaknesses on the recommend system which based on single model, such as content-based filtering and collaborative filtering technology existing start-cold and the data sparse problems. In addition, recommend system which based on association rules exist extracting problems and low automation degree problems, all of this caused the recommendation inefficientIn view of the above analysis and study the Single-Model's defects, this thesis offer a mixed model which based on content-based filtering and collaborative filtering and Web Log Analysis technology. This mixed model can select different model from person to person. It can make use of different recommendation model's advantage and as far as possible to improve the accuracy of the recommendation. Traditional recommend technology usually depend on the user's explicit information, which often caused the data shortages, and this thesis proposes a method which combining explicit and implicit information and using data mining technology to extracting user's data, and build user's interesting model. The implicit data mainly depend on the user's WEB logs. The improvement of two aspects can enhance the system's accuracy rating.Finally, using the improved recommendation model which based on hybrid model designs a personalized music recommendation system. The system provides different recommendation models, it can provide search services and recommendation services for public users and also can provide personalized recommendation services for registered users. So it can meet the requirements of the customers from different levels.Experimental data shows that the MAE's value which based on hybrid model is smaller than the recommendation system which based on single model, thus we could deduce that the improved system greatly improve the accuracy of the recommendation.