https://github.com/bflashcp3f/schema-to-json

Schema-Driven Information Extraction from Heterogeneous Tables
https://github.com/bflashcp3f/schema-to-json

Last synced: about 1 month ago
JSON representation

Schema-Driven Information Extraction from Heterogeneous Tables

• Host: GitHub
• URL: https://github.com/bflashcp3f/schema-to-json
• Owner: bflashcp3f
• License: gpl-3.0
• Created: 2023-05-23T04:30:11.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2024-04-01T02:03:49.000Z (9 months ago)
• Last Synced: 2024-04-02T02:42:38.397Z (9 months ago)
• Language: Python
• Homepage: https://arxiv.org/abs/2305.14336
• Size: 11.6 MB
• Stars: 17
• Watchers: 1
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• jimsghstars - bflashcp3f/schema-to-json - EMNLP 2024 Findings "Schema-Driven Information Extraction from Heterogeneous Tables" (Python)

README

        
Schema-Driven Information Extraction from Heterogeneous Tables

===============================================================================

This repo contains code and data associated with the paper ["Schema-Driven Information Extraction from Heterogeneous Tables"](https://arxiv.org/abs/2305.14336).

```

@misc{bai2023schemadriven,

    title={Schema-Driven Information Extraction from Heterogeneous Tables},

    author={Fan Bai and Junmo Kang and Gabriel Stanovsky and Dayne Freitag and Alan Ritter},

    year={2023},

    eprint={2305.14336},

    archivePrefix={arXiv},

    primaryClass={cs.CL}

}

```

# Task: Schema-to-JSON



# Installment

1. Create conda environment.

```

git clone https://github.com/bflashcp3f/schema-to-json.git

cd schema-to-json

conda env create -f environment.yml

conda activate s2j

```

2. Set up OpenAI API key with the environment variable `OPENAI_API_KEY`. If you want to use Azure, set up the environment variable `AZURE_API_KEY`.

3. Install from the source

```

pip install -e .

```

# Data

Four datasets (MlTables, ChemTables, DiSCoMat and SWDE) in our benchmark are available under the `data` directory.

# Experiments

Below are the commands to reproduce paper results. Make sure you set up `API_SOURCE` (`openai` or `azure`) and `BACKEND` (model name) in the script. For open-source models, use scripts with suffix `_os.sh`.

## MlTables

```

# Prompt (w/ error recovery)

bash scripts/mltables/prompt_error_recovery.sh

# Evaluation

bash scripts/mltables/eval.sh

```

## ChemTables

```

# Prompt (w/ error recovery)

bash scripts/chemtables/prompt_error_recovery.sh

# Evaluation

bash scripts/chemtables/eval.sh

```

## DiSCoMat

```

# Prompt

bash scripts/discomat/prompt_error_recovery.sh

# Evaluation

bash scripts/discomat/eval.sh

```

## SWDE

```

# Prompt

bash scripts/swde/prompt.sh

# Evaluation

bash scripts/swde/eval.sh

```