Research Article
An Approach for Item Recommendation Using Deep Neural Network Combined with the Bayesian Personalized Ranking
@INPROCEEDINGS{10.1007/978-3-030-30146-0_11, author={Zhongqin Bi and Siming Zhou and Xiaoxian Yang and Ping Zhou and Jiale Wu}, title={An Approach for Item Recommendation Using Deep Neural Network Combined with the Bayesian Personalized Ranking}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 15th EAI International Conference, CollaborateCom 2019, London, UK, August 19-22, 2019, Proceedings}, proceedings_a={COLLABORATECOM}, year={2019}, month={8}, keywords={Recommendation Stack Denoising Auto-Encoder Bayesian Personalized Ranking Deep learning The sparseness of matrix}, doi={10.1007/978-3-030-30146-0_11} }- Zhongqin Bi
Siming Zhou
Xiaoxian Yang
Ping Zhou
Jiale Wu
Year: 2019
An Approach for Item Recommendation Using Deep Neural Network Combined with the Bayesian Personalized Ranking
COLLABORATECOM
Springer
DOI: 10.1007/978-3-030-30146-0_11
Abstract
This paper proposes a deep neural network model (SDAE-BPR) based on Stack Denoising Auto-Encoder and Bayesian Personalized Ranking for the problem of accurate product recommendation. First, we use the Stack Denoising Auto-Encoder (SDAE) as the input of the item’s rating data and obtain the hidden features after encoding. Second, the Bayesian personalized Ranking (BPR) method is used to learn the hidden feature vector of the corresponding item. This model can avoid the influence of the sparseness of the matrix. Therefore, this model achieves the effect of more accurate recommendations of items. Third, to reduce the cost of model training, a unique pre-training and fine-tuning strategy is proposed in the deep neural network. Finally, based on the Movielens 20M dataset, the results of the SDAE-BPR, a traditional item-based collaborative filtering model and a user-based collaborative filtering model are compared. It is shown that the SDAE-BPR has higher accuracy. This method improves the accuracy of parameter estimation and the efficiency of model training.