The Research And Application Of E-commerce Recommender System Based On Hybrid Collaborative Filtering

The Internet is developing at a breathtaking pace. It has exerted a greatinfluence on all aspects of people’s life and has changed people’s ways incommunication, information retrieval and commercial business. Nowadays,E-commerce has become one of the most promising application areas of the Internet.With the rapid development of e-commerce, shopping websites are filled with allkinds of merchandises. Millions of options make it more and more difficult for usersto select the items they are interested in. E-commerce recommender systems emergeat the right moment and proper conditions. Recommender systems can help users tofind the items they need quickly by listing a series of recommended items for everyspecified user. Researchers have put forward a number of recommendationtechniques. Collaborative Filtering is one of the most successful recommendationalgorithms, and it has been used in many applications. However, some of theshortcomings in this recommendation algorithm need to be improved. The mainwork of this thesis is as follows:Firstly, it improves the process of finding nearest neighbours in collaborativefiltering algorithm by partitioned matrix and min-heap, reducing the demand formemory.Secondly, it resolves the problem of data sparsity existing in both of the twoalgorithms by filling the original user-item rating matrix with combined methods. Totake full advantage of the two algorithms and overcome the disadvantages, itintegrates the two algorithms by introducing a weighting factor and puts forward aweighted hybrid collaborative filtering algorithm, which can make interpretable andnovel recommendations.Thirdly, it makes experiments and analyses on the dataset from MovieLens. Theresults of these experiments show that the mean absolute error of the weightedhybrid collaborative filtering algorithm is less than that of the user-basedcollaborative filtering algorithm and the item-based collaborative filtering algorithm.That is, the recommendation results by the weighted hybrid collaborative filteringalgorithm are more precise to some extent. At the same time, the coverage rate and diversity of the recommendations generated by the weighted hybrid collaborativefiltering algorithm is satisfactory.Lastly, it designs a recommender system for e-commerce with the weightedhybrid collaborative filtering algorithm based on user and item. It develops ashopping website and implements the main function modules using DIV, CSS,Spring and Hibernate and applies the recommender system for e-commerce.