Vbpr:visual Bayesian personalized Ranking from implicit feedback-aaai2016-20160422

1, information

publication:aaai2016

2. What

Based on the improvement of BPR model: In the study of the partial order of commodity preference, the visual information of the product picture is added to the problem of cold start.

3. Dataset

Amazon Women,amazon Man,amazon phone,tradsy.com

4. How

Input

Ds (U,I,J): A collection of user-purchased partial-order relationship pairs, FI: item image eigenvector with deep CNN training

Output

VBPR model parameters.

Only MF models are used in this article

Mf:x=wh '. The output is w,h, and the parameter of the item picture embeding

Method: Randomly selected to purchase the group of item I, according to (U,i) > (u,j), random gradient descent method training model.

5, EVALUATION:AUC

BASELINE:RANDOM,POPRANK,MM-MF, BPR

6, conclusion

Thesis contribution: Improve the model, improve the experimental results.

Vbpr:visual Bayesian personalized Ranking from implicit feedback-aaai2016-20160422