MultiWOZ Helper

A code repository for creating and evaluating MultiWOZ data, with support for multiple versions (2.0, 2.1, 2.2, and 2.4). The scripts are based on the following repositories:

1. MultiWOZ Benchmark
2. MultiWOZ Evaluation

Install dependencies

python 3.12 or later

❱❱❱ pip install -r requirements.txt

Create data for Dialogue Sate Tracking (DST)

❱❱❱ python create_data.py --mwz_ver=<version>

Create delexicalized data

❱❱❱ python create_delex_data.py --mwz_ver=<version>

Evaluate

❱❱❱ python evaluate.py -v=<version> -m=<mode> -b -s -r -d -i=<input_file.json> -o=<output_file.json>

The default values for version and mode are 2.4 and test, respectively. The must include generated responses and/or belief states. For sample input files, refer to the predictions folder.