Research Of Recommendation Algorithm Based On Multi-ViewAnchor Graph Hashing

With the rapid development of the Internet and information technology, the amount of information and resource grows explosively, which makes it more difficult to obtain useful and valuable ones. As an effective tool to improve the ability of information acquisition for users, recommendation system has attracted lots of attention. Most of existing recommendation method based on collaborative filter only concerns user information from single view. Moreover, computations between high-dimensional vectors are always necessary in traditional methods, leading to high computational complexity and even infeasible for big data. In this thesis, we proposed a CF-based recommendation algorithm based on multi-view hashing learning to make use of information from multi views and avoid the computational bottleneck.The main contributions of this thesis are as follows:Firstly, we propose a method for multi-view anchor graph construction based on multi-view locally anchor embedding to fuse data from different views. Our proposed method exploits both view-specific and inter-view information to make an effective fusion of multi-view data.Secondly, based on the proposed multi-view anchor graph construction method, we proposed a hashing algorithm for multi-view data to learn similarity-preserved binary hashing code. We also propose an efficient method to calculate binary code for out-of-sample data.Finally, we propose an effective and efficient algorithm for collaborative filter recommendation in multi-view scenario. Utilizing the proposed hashing algorithm, we develop a fast similar user search method through approximate nearest neighbor search. Recommendation result can be calculated from data of similar users using user-based collaborative filter algorithm.