Study Of Personalized Recommendation Based On Evolutionary Multi-Objective Optimization And Deep Neural Networks

In this information explosion age, the recommender systems have stepped into public sight.The recommender systems are invented to help information selection and win popularity soon after. High popularity of the recommender systems demonstrates its great research value. However, the recommender systems' wide and successful application does not necessarily mean a happy ending. The exponential growths of data and users' higher even picky requirements have already brought about severe challenges to relevant studies on recommender systems.The rapid growth of Internet users has brought about increasing amount of user-generated data which causes the information overload. Although more user-generated data may yield more effective information to make the recommender system more efficient, too much useless tag data may face the system with the challenges of information redundancy and ambiguity. To solve this problem, the thesis comes up with an algorithm which uses deep neural network to extract the in-depth metadata from tag space layer by layer. The algorithm increases data analysis dimensions to make the metadata easy to be digged out from the tag space. These metadata are used to update user profile, thereby increasing the accuracy of conventional collaborative filtering algorithm. Experiments demonstrate the algorithm usability and its superior performance over the clustering-based recommendation algorithms.Besides, the impact of the number of the deep network layer on the algorithm performance is also studied.Fulfilling the various demands of multiple user group is another problem that should be paid special attention to. For a excellent personalized recommender system, both accuracy and diversity should be taken into account. In this case, the thesis builds a multi-objective recommendation model to optimize accuracy and diversity. The accuracy is evaluated by the probabilistic spreading method, while the diversity is measured by recommendation coverage. The proposed MOEA-based recommendation method can simultaneously provide multiple recommendations for multiple users in only one run. By comparison, the proposed algorithm can provide more diverse results for various user groups with higher accuracy.