External Knowledge Selection with Weighted Negative Sampling in Knowledge-grounded Task-oriented Dialogue Systems

Implements the model described in the following paper External Knowledge Selection with Weighted Negative Sampling in Knowledge-grounded Task-oriented Dialogue Systems in DSTC10_track2_task2 2022.

@misc{https://doi.org/10.48550/arxiv.2209.02251,
  doi = {10.48550/ARXIV.2209.02251},
  url = {https://arxiv.org/abs/2209.02251},
  author = {Han, Janghoon and Shin, Joongbo and Song, Hosung and Jo, Hyunjik and Kim, Gyeonghun and Kim, Yireun and Choi, Stanley Jungkyu},
  keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {External Knowledge Selection with Weighted Negative Sampling in Knowledge-grounded Task-oriented Dialogue Systems},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}

Setup and Dependencies

This code is implemented using PyTorch v1.8.0, and provides out of the box support with CUDA 11.2 Anaconda is the recommended to set up this codebase.

# https://pytorch.org
conda install pytorch==1.12.1 cudatoolkit=11.4 
pip install -r requirements.txt

Preparing Data and Checkpoints

Model Checkpoints.

• post-trained checkpoint for generation, fine-tuned checkpoint for generation

Data for training

• Synthetic training data

---

Automatic Data Construction for training

please refer data_processing/make_dstc10/make_synthetic_dstc10.py for data construction

training and evaluation

sh paper_(run/train)_(task_name).sh