Extractomat: Automatic Term Extraction (ATE) for English/German and Ukrainian

Features

1. Works on top of Spacy
2. Implements basic/combo-basic/c-value algorithms with extra flexibility and support for single word terms (see matcha.py)
3. Implements optional reranking using sentence transformers to weight the terms in the context of the document (see sbert_reranker.py)
4. Allows to run term extraction on txt/pdf/docx documents (see runner.py)
5. Comes with OTRT dataset (in English/German/Ukrainian)
6. Covered with tests.
7. Equipped with other experimental features (keybert_extract.py, gliner_extract.py) and scripts for measuring the performance on the ORTR dataset (tester.py, gt_verificator.py)

Installation

# clone the repo
git clone https://github.com/lang-uk/extractomat
cd extractomat

# Activate virtual environment and install dependencies
virtualenv venv
source venv/bin/activate
pip install -r requirements.txt

# Download Spacy models for your language of interest
spacy download uk_core_news_trf
spacy download en_core_web_trf
spacy download de_dep_news_trf

Running in the batch mode:

To run extractomat on the list of files, you can do:

python runner.py "my_papers_corpus/paper_*" --method cvalue --allow-single-word --n-max 6

Please consult with python runner.py --help for extra options.

Running tests:

Just start python -m pytest and relax.

OTRT dataset

The OTRT dataset (first page of the ONTOLOGIES OF TIME: REVIEW AND TRENDS paper) is in the otrt folder.

gt_terms_*.csv is the unique list of terms from the paper. gt_terms_*_full_ordered.csv is the complete list of terms in the correct order (as they occur in the text) TimeOnto Sample *.docx is the original text of the paper.

Practical application

We used extractomat in our experiments on building unsupervised bilingual glossary. See https://github.com/lang-uk/schmezaurus for details.

Citing the paper

Extractomat has been released as a part of the research results leading to the following paper:

Chaplynskyi, D., Wittenborg, T., Kosa, V., Rabby, G., Ignatenko, O., Auer, S., Ermolayev, V. (2025) Building Multilingual Terminological Bridges between Language-Specific Knowledge Silos. In: Proc. ICTERI-2025, CCIS, Springer (to appear)