Collaborative Filtering Recommendation Rersonalized Of Advertisement Based On The Mailbox

With the explosive growth of information resources, the "overload problem" is getting worse, helping users to find the desired quickly has been the focus of study, in which recommendation system has played an important role. E-mail platform is a necessary tool for everyday life, and contains lots of the user’s personal characteristics. In the domestic mail system platform, the study of personalized recommendation system for advertising is still insufficient.According to the characteristics mailbox platform, we proposed a recommendation system for mail platform-based on user-based and Item-based hybrid collaborative filtering algorithm. We analyse the user e-mail messages, generate user interest feature matrix by the Chinese word segmentation methods, grab the book data as the recommended ad database, computing users neighbor set and neighbor set items according to the user-project evaluation matrix, then generate the predicted recommended score for every user. The main work and innovation of this paper are as follows:1. This paper studies the user-based collaborative filtering algorithm and item-based collaborative filtering algorithm, propose a hybrid collaborative filtering algorithms and frameworks based on mail platform in order to solove the shortcoming of the traditional collaborative filtering algorithm. Besides, we propose a new algorithm to solve the cold start problem, which uses word2vec algorithm to do keyword expansion.2. Base on the email platform, we propose a new user’s multidimensional similarity for user-based collaborative filtering algorithms:calculating the cosine similarity using the user interest feature matrix to generate user similarity matrix, we consider the characteristics of social network of the mailbox users to seek user trustworthiness matrix, then obtain user similarity by user rating history, then getting the optimized user similarity by linear weighted all the three similarity matrix above.3. Propose a improved collaborative filtering-based algorithm:get the item-item similarity matrix by calculating the similarity of item characteristic vector, obtain the matrix which representative the target user’s interester for the project based on the user’s evaluation matrix history of articles. Then we get an improved collaborative filtering algorithm.4. We propose a user’s trust calculation method based on mailbox platform social network, and use the method in collaborative filtering algorithm, which improve the accuracy of the recommendation system; According to the rating characteristics of the mailbox system,we propose amendments to Jaccard correlation coefficient method to solve the similarity problem;The new algorithms we proposed have a very good performance in mailbox personalized recommendation, which made a good recommendation results in experiments.