How to construct a multi-lingual domain ontology

What it is about

The authors build a domain ontology, a DAG of concepts and containment relations, for organized crime. They first build it in English from Wikipedia (DBpedia, English edition of June 2012), and then for Hebrew. Several concept sources were tried. Contrasting corpora with TF-IDF and PMI reached 64.57% true positives on 80 expressions and were dropped; WordNet lacked most proper names and many links. Wikipedia article membership is a better source. Its naive reading, that every article under the category is relevant, held for only 174 of 319 sampled articles (55%). A parent-category vote at 80% gave an F-measure of 68%. A word-based classifier, trained on 290 abstracts and scoring each article by the ratio of relevant to irrelevant words, gave F between 0.96 and 0.99 as its certainty threshold varied; the price was 7.6% to 22.8% of articles left unclassified. Relations come from a depth-limited search over the category graph, choosing the most specific common ancestor by its number of children and its depth. For Hebrew, interlinks to English articles covered only 27 of 145 (19%), so the Hebrew concepts come from Hebrew categories filtered by whether an ancestor has an English counterpart; 43 Hebrew articles were accepted, and all were relevant on inspection. The authors could not measure recall.

Ideas

• I1. The word-relevance score as a feature for concept lists: each word gets the ratio of its counts in relevant versus irrelevant texts, and an article gets a score from its top words. Keep the unclassified band explicit instead of forcing a label. For Mova it must be deterministic, with the word tables in a BTreeMap.
• I2. A gold set for the Hebrew filter: about 150 Hebrew category pages, labelled for relevance, would measure the recall the authors could not. This is a held-out test for the counterpart-ancestor rule.
• I3. Take the 19% interlink coverage as a limit: a cross-lingual lexicon transfer through English links reaches only a small part of Hebrew. Modern Hebrew concept lists need Hebrew sources of their own.

For Mova

I1 is a feature for the concept lists in world, and it touches the Hebrew concept list in 8. Lexicon and register. I2 is a held-out test for that same list. I3 is a note for the planning in 8. Lexicon and register.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.63317/52p5qfw8sc9v

@inproceedings{chrizman2014how,
  title = {{How to construct a multi-lingual domain ontology}},
  author = {Chrizman, Nitsan and Itai, Alon},
  booktitle = {Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)},
  year = {2014},
  doi = {10.63317/52p5qfw8sc9v},
  url = {https://doi.org/10.63317/52p5qfw8sc9v}
}

References

12 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Biebow, Brigitte and Szulman, Sylvie. (1999). TERMI-NAE: a method and a tool to build a domain ontology. In EKAW ’99 Proceedings of the 11th European Workshop on Knowledge Acquisition, Modeling and Management, pages 49–66, London, UK. Springer-Verlag. find
2. de Melo, Gerard and Weikum, Gerhard. (2010). Menta: inducing multilingual taxonomies from Wikipedia. In Proceedings of the 19th ACM international conference on Information and knowledge management, CIKM ’10, pages 1099–1108, New York, NY, USA. ACM. find
3. Declerck, Thierry, Gómez-Pérez, Asunción, Vela, Ovidiu, Gantner, Zeno, and Manzano, David. (2006). Multilingual lexical semantic resources for ontology translation. In In Proceedings of the 5th International Conference on Language Resources and Evaluation. find
4. Gómez-Pérez, Asunción, Fernández-Lopez, Mariano, and Corcho, Oscar. (2010). Ontological Engineering. Springer. find
5. Hepp, Martin, Siorpaes, Katharina, and Bachlechner, Daniel. (2006). Harvesting Wiki consensus - using Wikipedia entries as ontology elements. In IEEE Internet Computing, pages 54–65. find
6. Lin, Feiyu and Krizhanovsky, Andrew. (2011). Multilingual ontology matching based on wiktionary data accessible via sparql endpoint. CoRR, abs/1109.0732. ↗
7. Maedche, A. and Staab, S. (2000). Semi-automatic engineering of ontologies from text. In In Proceedings of the 12th Internal Conference on Software and Knowledge Engineering, pages 231–239. find
8. Navigli, Roberto and Velardi, Paola. (2004). Learning domain ontologies from document warehouses and dedicated web sites. Comput. Linguist., 30(2):151–179, June. find
9. Suchanek, Fabian M., Kasneci, Gjergji, and Weikum, Gerhard. (2008). Yago: A large ontology from wikipedia and wordnet. Web Semant., 6(3):203–217, September. find
10. Syed, Zareen, Finin, Tim, and Joshi, Anupam. (2008). Wikipedia as an ontology for describing documents. In Proceedings of the Second International Conference on Weblogs and Social Media. AAAI Press, March. find
11. Velardi, Paola, Missikoff, Michele, and Basili, Roberto. (2001). Identification of relevant terms to support the construction of domain ontologies. In Proceedings of the workshop on Human Language Technology and Knowledge Management - Volume 2001, HLTKM ’01, pages 5:1–5:8, Stroudsburg, PA, USA. Association for Computational Linguistics. find
12. Wu, Fei and Weld, Daniel S. (2008). Automatically refining the Wikipedia infobox ontology. In Proceedings of the 17th international conference on World Wide Web, WWW ’08, pages 635–644, New York, NY, USA. ACM. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lingbulk-2014-chrizman-how-construct-multi-lingual-domain/