NAIC Entity Matching Demonstrator

Applying Prompt Engineering for Entity Matching

This repository contains the code for experimenting with various prompt engineering techniques for entity matching.

Setup

1. Clone the repository
2. Install the required packages using the following command:

    pip install -r requirements.txt

3. Download the datasets

    python utils/download_datasets.py

4. Ensure that OPENAI_API_KEY is set in the environment variables.

Notebook

The notebook Demonstrator.ipynb contains examples for how to run some experiments and visualize the results.

Running experiments

To run a zero-shot experiment on the abt-buy dataset using the natural prompt, 200 pairs, GPT-4o as the language model, context and "Be lenient in your judgement" added, run the following command:

    python main.py -d abt-buy -pf natural -n 200 -k 0 -llm gpt-4o -imp leneint -ctx

To run a few-shot experiment on the dblp_gs_dirty dataset using the tabular prompt, all pairs (if the number is larger than the dataset, the entire dataset will be used), 10 examples in the prompt with a 30/70 split of positive and negative examples respectively, GPT-3 as the language model, basic prompt format, and no sublte context, the following command can be used:

    python main.py -d dblp_gs_dirty -pf tabular -n 10000 -k 10 -llm gpt-3 -imp basic -ctp '(P,N)' -pn "30/70"

To get a list of all the available options, run the following command:

    python main.py --help