HeroNet

Codes are publicly available for dasfaa2023 manuscript ->

HeroNet: A Hybrid Retrieval-Generation Network for Conversational Bots

Requirements

absl-py==1.0.0 datasets==1.18.3 huggingface-hub==0.2.1 nltk rouge-score==0.0.4 sacrebleu==1.5.1 sentencepiece==0.1.95 tokenizers==0.10.3 torch==1.12.1 transformers==4.12.3

Usage

The hyperparameters for the HeroNet can be found in ''training_args_gan.py''. You can adjust them in the GANTrainingArguments class from the ''TrainGAN.py''.

If you want to run the model, use the command:

python TrainGAN.py

Dataset

We only provide datasets for demonstration, to access the full dataset, please refer to the paper A Large-Scale Corpus for Conversation Disentanglement

Output

The output of our model is presented in the ''result'' directory