Advanced Hybrid Information Processing. First International Conference, ADHIP 2017, Harbin, China, July 17–18, 2017, Proceedings

Research Article

Nearest-Neighbor Restricted Boltzmann Machine for Collaborative Filtering Algorithm

Download
237 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-73317-3_45,
        author={Xiaodong Qian and Guoliang Liu},
        title={Nearest-Neighbor Restricted Boltzmann Machine for Collaborative Filtering Algorithm},
        proceedings={Advanced Hybrid Information Processing. First International Conference, ADHIP 2017, Harbin, China, July 17--18, 2017, Proceedings},
        proceedings_a={ADHIP},
        year={2018},
        month={2},
        keywords={Restricted Boltzmann Machine Nearest neighbor Collaborative filtering Accuracy Over-fitting},
        doi={10.1007/978-3-319-73317-3_45}
    }
    
  • Xiaodong Qian
    Guoliang Liu
    Year: 2018
    Nearest-Neighbor Restricted Boltzmann Machine for Collaborative Filtering Algorithm
    ADHIP
    Springer
    DOI: 10.1007/978-3-319-73317-3_45
Xiaodong Qian1,*, Guoliang Liu1,*
  • 1: Lanzhou Jiaotong University
*Contact email: qianxd@mail.lzjtu.cn, 1185169269@qq.com

Abstract

Based on the restricted Boltzmann machine (RBM) collaborative filtering algorithm in recommendation phase easy to weaken the needs of individual users, and the model has poor ability of anti over-fitting. In this paper, the traditional nearest neighbor algorithm is introduced into the recommendation stage of RBM, use the characteristics of interest similarity, the nearest neighbor’s interest is used as the target user’s, strengthen the individual needs of users: First, using the traditional K-mean algorithm to find out the user’s n nearest neighbors; Then, using nearest neighbor to calculate the probability of users rating grades for the non rating items; Finally, weighted average score probability to the RBM model in the process of recommendation. Using benchmark data set Movielens experimental results show that the improved RBM model with nearest neighbor can not only improve the accuracy of the model results, but also increase the ability to resist over-fitting.