This is the accompany python package for the ACM MM 2021 paper：Empathetic and Emotionally Positive Conversation Systems with a Chatbot Emotion Generator (https://www.)
- python==3.7
- pytorch==1.5.1
- torch==1.5.1
- tqdm==4.48.0
- tensorboard==1.14.0
- texar-pytorch==0.1.1
- transformers==2.11.0
- sklearn
Edit the input,output_dict,output_src_tgt,output_src,output_tgt in build_vocab.sh, then run the following shell command to build dict for your dataset.
sh data/build_data.sh
The ${output_dict} file is the vocab you need for this dataset.
Run run.sh script file to start training, where mark means the personlized mark you set for this experiment (any string you want).
sh run.sh mark
After this shell command, you will see the generated log file (log_${mark} and err_${mark}) in logs/ and checkpoints in outputs/.
Run generate.sh script file to start testing (inference/prediction), where mark means the personlized mark you set when training, and pred_mark means the personlized mark you set for this testing (any string you want).
sh generate.sh mark epoch_id pred_mark
After this shell command, you will see the generated log file (logg_${mark} and errg_${mark}) in loggs/ and Bleu/Dist file in results/.
If you want to see both Bleu/Dist and sentiment results after running one script, then you can run the following shell command, where mark means the personlized mark you set when training, top20 means TopKSampling during inference where K=20, pred_mark is the mark you set for the pretrained sentiment classifier.
sh generate_multi_epoches_and_test_sentiment_scores.sh mark top20 weibo_senti_100k
After this shell command, you will see the generated log file (logg_${mark} and errg_${mark}) in loggs/ and Bleu/Dist file in results/, also you can find sentiment files in Bert/pred_labels/.