This repository contains the code for the Attention-based Ontology Alignment system, VeeAlign.

Please run the following commands:

1. conda create --name py37 python=3.7
2. pip3 install -r requirements.txt

Five datasets are made available with this repository, in the format required by the code:

1. Conference Dataset
2. Lebensmittel Dataset
3. Freizeit Dataset
4. Web directory Dataset
5. Multifarm Dataset

To add a new dataset:

1. Create a folder <dataset_name> inside the datasets directory.
2. This directory should contain two subdirectories ontologies and alignments, which contain the ontologies being aligned and the alignments respectively.

There are no naming conventions wrt the naming of the ontology/alignment files, but it is assumed that the RDF alignment file contains <onto1> and <onto2> tags denoting the ontologies being aligned.

To train a model, run python3 train.py.

To test a model, run python3 test.py <ontology1> <ontology2>.

Note that:

1. Both train.py and test.py will run with configurational parameters described in config.ini.
2. train.py will train on alignments specified in config.ini
3. test.py will load pretrained model specified in config.ini

Details on the configurational parameters and how to set them for reproducibility are described here.

There are five main script files:

1. src/train.py: This is the script that needs to be run to train your model.
2. src/test.py: This is the script that needs to be run to test your model.
3. src/data_preprocessing.py: This script contains code to preprocess data for running.
4. src/ontology.py: This script contains code to parse the ontology.
5. src/config.ini: This file contains the configurational values for main.py, and can be adjusted to make the code run for conference, lebensmittel, freizeit and web-directorydatasets. This is explained in further detail below.

We also provide the datasets used for experimentation, in the datasets folder (also included in the data appendix).

This section details in tabular format, the various configurational fields in config.ini and the best performing values for each of the datasets.