Collaborative Filtering Recommendation System Improved Algorithm

With the rapid development of Internet and information technology, the exponential growth of information has attract a lot of concern these day. However, to those who provide and those who receive information, this is a big challenge. How to highlight the information is a great challenge to information provider, especially to make innovation. When our user face the thousands of information sources, how to quickly and accurately find useful information on their own is the focus in today’s society research.The rapid growth of information resources while the utilization of information is gradually reduced,.for our user, the difficulty to pick up useful information in the sea of the information also get growth. To solve this problem, people use search engine in the early time. but search engine can only search the keywords with the same or similar information and make a certain degree of information filtering, but lack of personalized search results. On this condition, recommendation system emerged and remedy the shortcomings of search engines, it is based on the keywords entered by the user to get information, also include their social circle, as well as search history record, and make personalized recommendation service which establish a long-term and stable user intercourse relationship,not only improve customer loyalty, but also create a good and reliable information platform for the producers to achieve a win-win goal.This article by doing research on the recommendation system through the traditional algorithm analysis and research to understand the various algorithms recommendation systems. Existing user-based collaborative filtering algorithms and project-based collaborative filtering algorithm, although has been generally recognized. But there are shortcomings, such as the user similarity calculation error is large; there a user preference User rating issue. Meanwhile collaborative filtering algorithms are considered to contact the project resource itself has user ratings, but have not considered the level of trust between users and other issues, to solve these problems and improve the article carried.The main innovation of the article is as follows:1. By comparing the similarity algorithm, the traditional recommendation algorithm cosine similarity to the use of the modified cosine similarity, avoiding the problem of user preferences;2. Using a mixture of collaborative filtering algorithm, combined with the user rated similarity and the user-Project Properties similarity, nearest neighbor set to get more accurate;3. Based on the second step, by calculating the degree of trust between users, predicting the results obtained closer to the user’s true evaluation.