基于BP神经网络的协作过滤推荐算法

研究、探讨了协同推荐问题，提出了一种基于两层面的多个后向传播(BP)神经网络的协作过滤推荐算法(TMNN-CFRA). 两层面的多个BP神经网络协同工作，高层面BP网反向误差传播直至低层面多个人工神经网络(ANN)进行网络权值修正，以此为基础，借助用户评价等特征前向给出项目推荐. 标准评测集Movielens上的实验评测表明了TMNN-CFRA的可行性和有效性.

BP Neural Networks-Based Collaborative Filtering Recommendation Algorithm

As an effective method for resource recommendation, traditional collaborative filtering algorithms are extensively adopted on web. However, collaborative filtering method has potential problems such as that the similarities of nearest neighbors are not as high as expected and consequently, the recommendation precision is not high enough. In our paper, we give a novel method for similarity measurement and item prediction by taking advantage of multiple BP networks cooperating together. Specifically, higher layer BP conversely modifies the multi-similarity presented by lower layer BP networks, and based on which, better recommendation performance is acquired in the process of forward propagation. Experiment results on Movielens dataset show that our method is feasible and effective on item recommendation.