Research And Implementation Of Recommender System Based On Collaborative Filtering Model And LFM

With the rapid development of E-Commerce, more and more customers are accustomedto purchase goods by E-Commerce. However, E-Commerce sites provide so massivecommodities that the customers are difficult to make quick decisions, and the customers takelots of time and energy to seek out the merchandises which are satisfied with their demands. Itis difficult and popular for E-Commerce research to build an effective mechanism, whichhelps people to reduce time and energy and ensure the quality of acquire the information.Recommended system provides users with a hassle-free experience, users no longer need tospend a lot of time for searching useful information, but can focus on their real interest andneeded goods or information.In this thesis, Recommender System based on collaborative filtering and LFM areresearched through theoretical research and practical application. There are three key pointsdiscussed: comparative study of collaborative filtering recommender algorithm model,optimization of recommender algorithm through the method of combining LFM model withneighborhood collaborative filtering model, and the design of the recommendation systembased on integration model. Proposed an optimized algorithm and then demonstrated thecorrectness of the algorithm by comparative experiments on the Netflix dataset.Firstly, according to the trend development of the personalized recommendation system,the most popular variety of algorithms, which including collaborative filteringrecommendation system, as well as the mechanism and theory of neighborhood-basedcollaborative filtering algorithm, are laid special stress on studying. Three recommended stepsusing collaborative filtering algorithm are analyzed：building scoring matrices, similaritycalculation and prediction and recommendation. Compared the influence on the accuracy ofrecommendation based on different similarity metrics, as well as the performance of differentcollaborative filtering algorithm in the data set, the thesis summarizes the advantages anddisadvantages of each recommender algorithm, through experiments on Movielens dataset.Secondly, the principle of LFM model and the significant in the thesis are expounded.Also, the thesis improves the classical LFM model, and sums up the implementation steps ofthe model in the recommendation system. In addition, the characteristics and limitations ofrecommendation algorithm between the LFM model and the neighborhood collaborativefiltering model are compared.Thirdly, an optimized recommender algorithm is designed through the method of combining LFM model with neighborhood collaborative filtering model. Therefore, the thesisimproves the accuracy of the algorithm through merging the implicit feedback data in themodel. Then an experiment is carried out on Netflix data set, the results proved that the newalgorithm is superior to the traditional collaborative filtering algorithm.Finally, a recommender system prototype, which is based on the open-source platformstructures, is built. What’s more, the database structure and the system’s overall function areexplained. The design of recommender system prototype based on Mahout Developmentplatform, which achieved the purpose of a personalized intelligent recommendation, andverified the validity of the method.