Research On Personalized Recommendation Algorithm Based On Collaborative Filtering

With the in-depth researches of Internet technology and the rapid developments of Internet applications,the global data is growing at an exponential rate,the information overload is more and more serious.In order to use these Internet data more effectively,recommended system has been widely used in the fields of online-shopping,audio and video,e-news and magazine,information retrieval,news and so on.The merits and demerits of the recommendation of an Internet application or an Internet products,may even affect the user's product sticky.As a core component of the recommendation system,it is self-evident that the recommended algorithm plays a decisive role throughout the system.As a simple and efficient algorithm,collaborative filtering technology has become the most major and the most successful recommendation algorithm,but this technology still has many problems need to be solved,such as: user sparseness caused by the recommended performance dropped significantly,cold start problems and the traditional algorithm overlooking the users' changing preferences and other issues.The author deeply studies and researches the traditional collaborative filtering technology in this thesis,including the traditional user-based and item-based collaborative filtering technology,a user-and-item-based hybrid collaborative filtering technology and a collaborative filtering technology which is based on the gradual forgetting.Based on the forefathers' study,the author has carried out a more in-depth study about this technology,the following is the main idea of this thesis:1.This thesis introduces the background of recommendation system,the technology of data mining and information filtering related to individual recommendation,as well as analyze the recommended process and the merits and demerits of several common-used collaborative filtering algorithms.Then the author analyzes the different recommendation technologies.And finally,it discusses the existing issues of the applications at present,and paves the way for the next step of the improved algorithm experiment.2.The traditional collaborative filtering technologies are analyzed and studied,the corresponding projects are carried out,and the data are retained.3.A hybrid collaborative filtering improvement technique based on user-item is proposed to reduce the influence of data sparseness on similarity calculation and compared with traditional collaborative filtering results,verified the improvement of the algorithm in the recommended results.4.Consider the effect of gradually forgetting on user ratings and study the role of forgetting curves in personalized recommendations and compared with traditional collaborative filtering results,verified the improvement of the algorithm in the recommended results.