Research On Key Technologies Of E-commerce Personalized Recommendation

With the popularity of Internet and wild applications of E-commerce, more and more information swirles in the net. Consumers enjoy the convenience of online shopping, on the other hand, they have been plagued with the problem of information overload. E-commerce Recommender Systems emerge in order to find the right information from the information sea.Collaborative filtering is one of the most successful technologies for personalized recommender systems. This thesis takes collaborative filtering algorithm as the research object from the angle of improving the quality of the active user's nearest neighbor. Traditional collaborative filtering algorithms exist the sparsity problem and do not consider the drift of user interests. To solve these problems, this thesis put forward an improved collaborative filtering algorithm. Compared to the traditional algorithm, the algorithm proposed in this thesis has the following advantages: The first, to reduce the sparsity of the dataset, implicit rating approach is used to expand rating matrix for those items which have not been rated. Secondly, this thesis advanced a similarity algorithm on the basis of time weight and positive interests of active user. The similarity of two users is calculated only based on those items which the active user has have a positive interest, at the same time, the more later rated items are given the higher weight. So the algorithm can effectively reflect the drifts of the active user's interest and then improve the quality of the active user's nearest neighbor. The last, to find more"true"neighbors, the algorithm uses the original item classification model of E-commerce as a standard classification model and find the nearest neighbor of active user in different sub-categories.An experiment is designed to validate the effect of the proposed algorithm in this thesis. The data provided on MovieLens and EachMovie are used and the Mean Absolute Error is analysed in the experiment. The experiment results show that the improved algorithm has a higher accuracy than the traditional user-based collaborative filtering algorithm in predict rating.