GraphDR is an effective, efficient, and easy-to-deploy matching (i.e., candidate generation in two-stage recommendation) model based on heterogeneous interactions among user, item, category, tag, word, and media. The details of GraphDR are in Improving Accuracy and Diversity in Matching of Recommendation with Diversified Preference Network（TBD-2021) Ruobing Xie*, Qi Liu*, Shukai Liu, Ziwei Zhang, Peng Cui, Bo Zhang and Leyu Lin.

python 2.7.15 tensorflow 1.14.0

sh -x train.sh

If the codes help you, please cite the following paper:

Ruobing Xie*, Qi Liu*, Shukai Liu, Ziwei Zhang, Peng Cui, Bo Zhang and Leyu Lin. Improving Accuracy and Diversity in Matching of Recommendation with Diversified Preference Network（TBD-2021). (* indicates equal contribution).