Research On Personalized Recommendation Algorithm Based On Active Learning

Along with the rapid development of computer and internet technologies, redundant information and useless information grow quickly, which has caused serious information overload problem. In order to solve this problem, grouping methods, search systems and many other techniques have been proposed. Recommender system has been one of the key technologies for information overload problem. The objective of recommender system is to dig out the potential interest of users and help to complete information filtering.In this thesis, we introduce active learning to recommender system. Specifically, we adopt active learning techniques to solve the new users’ cold start problem in recommender system. Experimental results on MovieLens datasets and MovieRating datasets show the effectiveness of our methods.The main contributions of this thesis are:(1) This thesis analyzes the mainstream algorithms of recommender system and active learning to date with a focus on Baseline SVD algorithms based on matrix factorization. Our empirical study shows that Baseline SVD has better performance than other baseline algorithms.(2) This thesis summarizes the difficulties and emphasis of recommender system. Aiming at the cold start problem for new users we propose a query strategy based on the shock factor of estimated score. Experimental results show that the proposed method accelerates the start of recommending for new users comparing with random query strategy and popular query strategy. We propose a better query strategy by combining the two methods.(3) The performance of the method based on the shock factor is greatly affected by the model and other factors. This thesis proposes another algorithm based on the error of assessment. Comparing with the former method, the second has better performance.(4) This thesis presents a hybrid recommender system to solve the sparse data problem by combining the Slope One algorithm and Baseline SVD algorithm. Experiments show that the accuracy of the proposed recommender algorithm is improved.