The source code for our WSDM 2022 Paper "C2-CRS: Coarse-to-Fine Contrastive Learning for Conversational Recommender System"
- Python == 3.8
- Pytorch == 1.8.1
- CRSLab == 0.1.2
We proposed a novel contrastive learning based coarse-to-fine pre-training approach for conversational recommender system. By utilizing the coarse-to-fine pre-training strategy, multi-type data representations can be effectively fused, such that the representations for limited conversation context are further enhanced, which finally improve the performance of CRS.
We use two datasets in our paper, all of which have been uploaded to Google Drive and Baidu Netdisk (password: 2ho6).
The downloaded dataset folder should be placed in the data folder.
We have trained our model on two datasets and saved the parameters, all of which have been uploaded to Google Drive and Baidu Netdisk (password: 44kr).
The downloaded save folder should be placed in the root folder of this project.
You can train the model.
sh script/redial/train/redial_rec_train.sh
sh script/redial/train/redial_conv_train.sh # remember to change --restore_path
sh script/tgredial/train/tgredial_rec_train.sh
sh script/tgredial/train/tgredial_conv_train.sh # remember to change --restore_path
You can also test the model has been saved by us.
sh script/redial/eval/redial_rec_eval.sh
sh script/redial/eval/redial_conv_eval.sh
sh script/tgredial/eval/tgredial_rec_eval.sh
sh script/tgredial/eval/tgredial_conv_eval.sh
If you have any questions for our paper or codes, please send an email to [email protected].
Our code is developed based on CRSLab
And thanks the code from SimCLR
Any scientific publications that use our codes and datasets should cite the following paper as the reference:
@inproceedings{10.1145/3488560.3498514,
title = {C²-CRS: Coarse-to-Fine Contrastive Learning for Conversational Recommender System},
author = {Zhou, Yuanhang and Zhou, Kun and Zhao, Wayne Xin and Wang, Cheng and Jiang, Peng and Hu, He},
booktitle = {WSDM},
year = {2022},
}