The implementation of Reinforced Negative Sampler for implicit recommender system.
RNS is a better negative sampler for recommendation with exposure data. This is our official implementation for the paper:
Jingtao Ding, Yuhan Quan, Xiangnan He, Yong Li, Depeng Jin, Reinforced Negative Sampling for Recommendation with Exposure Data, In Proceedings of IJCAI'19.
If you use the codes, please cite our paper. Thanks!
Data is the Zhihu dataset in the paper.
Please run code with shell in 'sh/'
RNS: bash rns.sh
KBGAN: bash kbgan.sh
DNS: bash dns.sh
We have provided a pretrained model file for above methods.
BPR: bash bpr.sh
ItemPop: bash itempop.sh
Two evaluation mode: topK or List (as the paper)