PoWareMatch: Matching as a Process and the Case of a Human Matcher

Prerequisites:

1. Anaconda 3
2. Pytorch

Getting Started

Installation:

1. Get human (algorithmic) matching data.
   1.1. An example dataset is available for download: Human Dataset Sample
   1.2. An example reference match is given in Reference Match Example
   1.3. An example algorithmic match is given in Algorithmic Match Example
2. Clone the PoWareMatch repository
3. Update Config with your configuration details.

Usage

1. RunFiles folder consists of several runnables to train and test your dataset using a 5-fold cross validation.
   1.1 The runnable files differ in the extent of which the experiments are run (e.g., use pred_y to only train and test a calibration model)
2. Once the models are trained, you can use the PoWareMatch inference to calibrate human matching and generate better matches.

The Paper

PoWareMatch: a Quality-aware Deep Learning Approach to Improve Human Schema Matching. Roee Shraga, Avigdor Gal, ACM Journal of Data and Information Quality (JDIQ), 14(3), 1-27, 2022.

BibTeX:

@article{shraga2022powarematch,
title={PoWareMatch: a Quality-aware Deep Learning Approach to Improve Human Schema Matching},
author={Shraga, Roee and Gal, Avigdor},
journal={ACM Journal of Data and Information Quality (JDIQ)},
volume={14},
number={3},
pages={1--27},
year={2022},
publisher={ACM New York, NY}
}

The Team

PoWareMatch was developed at the Technion - Israel Institute of Technology by Roee Shraga under the supervision of Prof. Avigdor Gal