Research On Collaborative Filtering Algorithm Based On Users' Real-time Feedback

With the popularity of internet and the rapid development of e-commerce, the users, who have to find the goods they needed in the vast amounts of products information, are facing more and more serious problem of information overload. Recommend system can resolve the problem effectively through interacting with customer, and recommend goods to them according their interest, which makes products more attractive and builds a steady relationship between customer and website. But the problem of accuracy and effectiveness has restricted the development of Recommend System.Currently collaborative filtering recommendation algorithm is the most widely used recommendation algorithm, which used mainly time on the calculation of the similarity. As the increase of users and items, recommended system needs to constantly re-calculating their similarity, which can't meet the needs of users. To some extent, model-based collaborative filtering can solves the difficult problem, while the long cycle of data update decrease the recommended accuracy. To solve the above problem, a collaborative filtering algorithm based on users' timely feedback is proposed, which achieves that recommender system can finish the real-time updating of the model data when a new rating is submitted by active user. Hence, recommender system can reflect the changing of user interest accurately. The main research work of this paper is as follows:1. We review the development of the recommender system and recommender algorithm, conclude the basic principles of collaborative filtering and its existing problems, so as to lay a foundation for the following work.2. In order to update the change of user's interest timely, this paper presents a recommender model based on real-time user feedback, which is able to receive the feedback of user online and partially update user similarity by its update mechanism. This model can be divided into two parts, one is the module of direct feedback and the other is the module of indirect feedback, both of which reflect the influence of user feedback from different angles.3. For this model, we also proposed an algorithm based on real-time user feedback, which achieve the goal of real-time updating data on the traditional collaborative filtering algorithms by changing the style of similarity formula. In addition, we also introduce the adjacency list structure in order to reduce the complexity of algorithm, which reach the purpose of updating model feature online.4. In the experiment, this three-part experiment was designed to test the algorithm results. In order to evaluate the performance of the algorithms, we not only use the traditional mean absolute error (MAE) and the average recommended time (MRT) of these two indicators, but also raise the average evaluation time (MAT). The experiments results indicate that the algorithm can improve the recommendation accuracy efficiently and reduce the recommendation time significantly. Finally, we discussed the future of the recommended system.