Research And Application Of Recommendation Algorithms

With the rapid development of Internet, people are increasingly dependent on information resources. Faced with such a growth of information on the Internet today, how can users find the information they need is a matter of concern. In order to solve the problem of information overload, many scientists and engineers have put forward tons of effective solutions, from which the recommendation system is the most representative.Because the website will provide a ranking list to recommend items, users can easily find popular items which they need, but users’ demand for non-popular(long-tailed) goods cannot be satisfied by the same rankings. Recommendation system can better explore the users’ demand for unpopular ones. Recommended system provides each user with a corresponding recommendation list, and users can get the items they are interested in via it, which includes not only popular items but long tail ones. In order to provide accurate recommendation list to users, it is necessary to fully study the interest of users.Recommendation algorithm is the core of recommendation system. There are many recommendation algorithms, including recommendation algorithm based on content,the recommendation algorithm based on association rules, and collaborative filtering algorithm. As the most widely used recommendation algorithm in the academia and industry now, collaborative filtering algorithm has been researched by scientists and engineers for a long time.In order to solve the problems of collaborative filtering algorithm, this thesis has done some research work. Firstly, in order to eliminate the influence of time effect, this thesis proposes a collaborative filtering algorithm T-UserCF which mixes time information to further improve the accuracy of collaborative filtering algorithm, and designs algorithm performance testing to verify the validity of it. Secondly, in order to solve the scalability in collaborative filtering algorithm, this thesis proposes an incremental collaborative filtering algorithm I-UserCF to effectively reduce the computation loading and the time overhead of recommendation algorithm without loss of information, and designs algorithm performance testing to verify the validity of it. Thirdly, in order to adapt to the massive data processing requirements in recommender system, this thesis presents a parallel I-UserCF algorithm based on MapReduce under cloud computing, and realizes this algorithm on the Hadoop platform. Lastly, this thesis completes a movie recommendation system prototype, and applies parallel I-UserCF algorithm to it.Experiments and application show the research findings of this thesis have good validity and practicality.