Research On Advertisement Recommendation Based On User Interests Collaborative Filtering Algorithm

With the rapid development of technology, the information from Internet is multifarious, it is a focus of academic and business how to advertising for users who with some specific interests, and improve the economical benefit. In this paper, the first job is analyze and mining data which recoded user’s network actions from X university communication website. Second job is improve calculate and recommendation mechanism of collaborative filtering algorithm, build mapping between user’s interests and AD recommendation method base on collaborative filtering algorithm, for accurate AD recommend to users with special interests.The mainly job of this paper including catch data source and analyze it, abstract user interest method, improve collaborative filtering algorithm and proving AD recommendation system.For more accurate user interest, the key is data source and analyze it. There are two type data source in this paper:static data and dynamic data; the static data source from enroll information on X university communication website, to make sure readability of this type data, this paper take a lot of measures to get it:reject invalid characters, symbols, modal particle; The dynamic data source catch from record log which recording user action, Internet web data caught by adaptive web similarity multithreaded topic crawler, and then split those data to words, in this part, paper use directed acyclic graph(DAG) and dynamic programming maximum probability map to make sure accurately of words, for a better efficiency of abstract user interest method use compressed Trie structure to storage words for matching is the next job.This paper use static and dynamic data to extract user interest method, think about the regularity of static data, use node of N tree structure to storage static data which by processed, the parent node and other node(not leaf node) use for storage more abstract interest information. By using relationship search algorithm to calculate user interest belong which interest set of N tree for extract use interests. After a series of analysis process on dynamic data, this paper extract user interests based on word frequency strategy.For more accurate of forecast result in predication period of collaborative filtering algorithm, this paper not only use the user click history what through user who enrolled in the site but add user interest method in this period to forecast user click-through rate. Use the regression calculate method to eliminate inertial behavior of user ratings for the result of accurate. Through make the data source diversification of collaborative filtering algorithm, the accurate of user click-through rate has been improved.For all of the above analysis work, this paper have designed a crawler software and recommendation system, and applied recommendation system in X university communication site. The experiment proved, the performance of the crawler software is better than other conventional crawler software, the web been crawled have a larger correlation with user; recommended products and user interest in closer.