Code for CRFR

The implementation of our paper accepted by EMNLP 2021: CRFR: Improving Conversational Recommender Systems via Flexible Fragments Reasoning on Knowledge Graphs

The code and data are developed based on KGSF. Due to time reasons, it has not been transformed into a standardized code framework, so it seems that the code is bad.

Environment

• python==3.6.12
• torch==1.3.0+cu100
• torch_geometric==1.3.2

Train steps

Pre-train

• python run.py --pre_train True

Train policy reasoning

• python run.py --train_reasoning True --learningrate 1e-4 --epoch 20

Train recommendation module

• python run.py --train_rec True --learningrate 4e-4

Train dialog module

• python run.py --is_finetune True

Citation

If you find our work useful for your research, please kindly cite our paper as follows:

@inproceedings{DBLP:conf/emnlp/ZhouWHH21,
  author       = {Jinfeng Zhou and
                  Bo Wang and
                  Ruifang He and
                  Yuexian Hou},
  editor       = {Marie{-}Francine Moens and
                  Xuanjing Huang and
                  Lucia Specia and
                  Scott Wen{-}tau Yih},
  title        = {{CRFR:} Improving Conversational Recommender Systems via Flexible
                  Fragments Reasoning on Knowledge Graphs},
  booktitle    = {Proceedings of the 2021 Conference on Empirical Methods in Natural
                  Language Processing, {EMNLP} 2021, Virtual Event / Punta Cana, Dominican
                  Republic, 7-11 November, 2021},
  pages        = {4324--4334},
  publisher    = {Association for Computational Linguistics},
  year         = {2021},
  url          = {https://doi.org/10.18653/v1/2021.emnlp-main.355},
  doi          = {10.18653/v1/2021.emnlp-main.355},
  timestamp    = {Thu, 16 Jun 2022 20:35:19 +0200},
  biburl       = {https://dblp.org/rec/conf/emnlp/ZhouWHH21.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}