We now publish the codes and the learned parameters of all models (baselines and ours) in our paper Towards Topic-Guided Conversational Recommender System, to appear in International Conference on Computational Linguistics, 2020 (COLING2020).
Our dataset has been published in TG-ReDial, which contains the description and dataset of our TG-ReDial.
Thanks to Yuanhang's hard work, in this project, you can implement the models in our paper by simple scripts. So please do not save your star :)
python==3.8.6
torch==1.6.0
pip install -r requirements.txt
This project only contains the code of TG-ReDial. You can get original data from GoogleDrive or BaiduNetDisk, the password for baidunetdisk is cc2o.
For the convenience of implementing these models, we have preprocessed all the data and prepared all trained model parameters, you can get them from GoogleDrive. You should download them and place them in proper path, then use our script to test or retrain the model.
First, placed the data_move.tar.gz in ./TG_CRS_Code/ Then, run the following code
tar zxvf data_move.tar.gz
bash data_move/move_data_back.sh
1. Ours
```
cd Recommender/Union
# training
bash script/train_Ours.sh
# testing
bash script/test_Ours.sh
```
2. BERT
```
cd Recommender/Union
# training
bash script/train_BERT.sh
# testing
bash script/test_BERT.sh
```
3. TextCNN
```
cd Recommender/TextCNN
# training
bash script/train.sh
# testing
bash script/test.sh
```
4. SASRec
```
cd Recommender/Union
# training
bash script/train_SASRec.sh
# testing
bash script/test_SASRec.sh
```
5. GRU4Rec
```
cd Recommender/GRU4Rec
# training
bash script/train.sh
# testing
bash script/test.sh
```
6. KBRD
```
cd Conversation/KBRD
bash scripts/both.sh <num_exps> <gpu_id>
```
7. ReDial
```
cd Conversation/KBRD
bash scripts/baseline.sh <num_exps> <gpu_id>
```
1. Ours
```
cd Conversation/Union
# Prepare the predicted data, note that we have prepared, so you can skip this step
# To run ours Response Generation Model, we need to use movie predicted by
# ours recommender model and topic predicted by ours topic prediction model.
# After train the latter two models, you can use this command to get the
# predicted consequence
bash ../../TopicGuiding/Ours/script/test.sh <gpu_id>
cp ../../TopicGuiding/Ours/data/identity2topicId.json data/data_Ours
bash ../../Recommender/Union/script/gen_pred_mids.sh
cp ../../Recommender/Union/data/data_p_Ours/identity2movieId.json data/data_Ours
# prepare for data, note that we have prepared, so you can skip this step
bash script/Ours/prepare_data.sh
# training
bash script/Ours/train.sh
# testing ppl
bash script/Ours/test_ppl.sh
# generating
bash script/Ours/generate.sh
# eval generation
bash script/Ours/test_gene_metric.sh generation/v11051_gen_output.txt
```
2. GPT2
```
cd Conversation/Union
# prepare for data, note we have prepared, so you can skip this step
bash script/GPT2/prepare_data.sh
# training
bash script/GPT2/train.sh
# testing ppl
bash script/GPT2/test_ppl.sh
# generating
bash script/GPT2/generate.sh
# eval generation
bash script/GPT2/test_gene_metric.sh generation/v1116_gpt2_gen_output.txt
```
3. Transformer
```
cd Recommender/Transformer
# training
bash script/Transformer/train.sh
# testing ppl and generating
bash script/Transformer/test_ppl.sh
# eval generation
bash script/test_gene_metric.sh output/output_test_both_epoch_-1.txt
```
4. KBRD
```
cd Recommender/KBRD
# training and testing ppl
bash scripts/t2t_rec_rgcn.sh <num_exps> <gpu_id>
# generating and eval generation
bash myscript/generate.sh
```
cd TopicGuiding/Model_You_Want
# training
bash script/train.sh
# testing
bash script/test.sh
If you use our code, please kindly cite our papers. Towards Topic-Guided Conversational Recommender System
@inproceedings{zhou2020topicguided,
title={Towards Topic-Guided Conversational Recommender System},
author={Kun Zhou and Yuanhang Zhou and Wayne Xin Zhao and Xiaoke Wang and Ji-Rong Wen},
booktitle = {Proceedings of the 28th International Conference on Computational
Linguistics, {COLING} 2020, Barcelona, Spain, December 8-11,
2020},
year = {2020}
}