Elamur Research⌕
☰
home / papers / 2012 / Mapping Arabic Wikipedia into the Named Entities Taxonomy

Mapping Arabic Wikipedia into the Named Entities Taxonomy

Paper · 2012 · Fahd Alotaibi and Mark Lee

Summary

The paper addresses a gap in Arabic named-entity resources: supervised NER needs labelled data beyond newswire, and Arabic Wikipedia (AW) could supply it if each article were linked to an entity type. The authors treat this as document classification. They adapt the ACE (2008) coarse taxonomy for open-domain text by adding Product and a Not-Named-Entity label. Two native speakers annotated 4,000 AW articles. The authors then compare a grid of three factors:

  • Feature representation: term presence, TF, TF-IDF.
  • Feature set: raw, filtered, stemmed, or enhanced with AMIRA tokenisation, POS filtering and location tags.
  • Classifier: Naïve Bayes, Multinomial NB, SVM, SGD.

Instead of whole articles, they use only four high-signal zones: the title, the first sentence, the category links and the infobox parameters. The best combination, TF-IDF with filtered features and SGD, reaches about 90% precision, recall and F-measure. This shows that AW articles can be mapped automatically to NE classes with reasonable accuracy. Such a mapping can then serve as a basis for building an Arabic NE corpus from Wikipedia.

Key points

  • The scheme has nine labels: seven come from ACE (2008), namely PER (person), ORG (organisation), LOC (location), GPE (geo-political entity), FAC (facility), VEH (vehicle) and WEA (weapon), and two were added, PRO (product) and NOT (not a named entity). About 74% of the annotated articles turned out to denote named entities.
  • Annotation ran in three cycles (500, 2,000 and 4,000 articles), with per-class Kappa reported after each cycle.
    • FAC was the weakest class early on (54) and improved to 96 by the end.
    • All final per-class scores fall between 96 and 100.
    • The dataset is freely released.
  • Term presence performs poorly: Naïve Bayes peaks at F 0.70 with the enhanced feature set, versus 0.56 with raw features. TF and TF-IDF do much better, for example MNB with enhanced features reaches 0.88 under TF.
  • Shallow stemming (NLTK ISRIStemmer) brought no gains. AMIRA clitic tokenisation combined with keeping only nouns did help.
  • Enhanced language-dependent features were not uniformly best, particularly in combination with TF-IDF. Experiments used an 80/20 split, scikit-learn and NLTK.

Interesting ideas

  • I1. Each token is prefixed with the zone it came from (title, first sentence, category, infobox), so the same word in different fields counts as a separate feature; the authors found this surprisingly effective.
  • I2. Classification uses only a few compact, highly informative parts of a long, variable-length document instead of the full text, which cuts noise and runtime.
  • I3. The label set includes an explicit "not a named entity" class rather than forcing every item into some type, and the taxonomy was extended with Product after inspecting the real domain.
  • I4. Agreement is measured per class over successive annotation rounds, starting with a training batch, which exposes weak categories like FAC before scaling up.
  • I5. Every number is mapped to a digit-shape string (a year becomes "dddd"), so features generalise over specific values.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (world): when typing entities from Wikidata/WordNet records, keep field provenance (label vs. gloss vs. class link) as part of the evidence so the type decision can say which field drove it.
  • I2 (world): when compiling the ontology, derive entity types from glosses' first sentence plus category/hypernym links rather than whole descriptions, keeping the reasoning short and traceable.
  • I3 (world): the entity layer should carry an explicit NOT/none type and a domain-driven extra class list, so the world model can refuse to reify common nouns as named entities.
  • I4 (learn): track per-label agreement between witness parsers across silver-data rounds to spot relations or types where guidelines or repair rules need work.
  • I5 (tokens): add shape features for numerals and dates in the tagger, so year, quantity and ordinal patterns are learned without memorising individual numbers.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of BirminghamBirmingham · United Kingdom25 works · 62,564 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Fahd Alotaibi and Mark Lee. 2012. Mapping Arabic Wikipedia into the Named Entities Taxonomy. Proceedings of COLING 2012: Posters.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{alotaibi2012mapping,
  title = {{Mapping Arabic Wikipedia into the Named Entities Taxonomy}},
  author = {Fahd Alotaibi and Mark Lee},
  booktitle = {Proceedings of COLING 2012: Posters},
  year = {2012},
  url = {https://aclanthology.org/C12-2005/}
}

References

26 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. AbdelRahman, S., Elarnaoty, M., Magdy, M., and Fahmy, A. (2010). Integrated machine learning techniques for Arabic named entity recognition. IJCSI International Journal of Computer Science, 7(4):27–36. find
  2. Abuleil, S. (2004). Extracting names from Arabic text for question-answering systems. In Proceedings of Coupling Approaches, Coupling Media and Coupling Languages for Information Retrieval (RIAO 2004), pages 638–647, Avignon, France. find
  3. ACE (2008). Ace (automatic content extraction) English annotation guidelines for entities. [accessed 12 June 2012]. find
  4. Alotaibi, F. and Lee, M. (2012). Using Wikipedia as a resource for Arabic named entity recognition. pages 27–34, Rabat, Morocco. In Proceeding of the 4th International Conference on Arabic Language Processing (CITALA12) . find
  5. Attia, M., Toral, A., Tounsi, L., Monachini, M., and van Genabith, J. (2010). An automatically built named entity lexicon for Arabic. In Proceedings of the Seventh conference on International Language Resources and Evaluation (LREC10), Valletta, Malta. European Language Resources Association (ELRA).
  6. Benajiba, Y., Diab, M., and Rosso, P. (2008). Arabic named entity recognition: An svm-based approach. In Proceedings of 2008 Arab International Conference on Information Technology (ACIT), pages 16–18, Amman, Jordan. Association of Arab Universities. ↗
  7. Benajiba, Y. and Rosso, P. (2007). Anersys 2.0: Conquering the NER task for the Arabic language by combining the maximum entropy with POS-tag information. In Proceedings of the Workshop on Natural Language-Independent Engineering, 3rd Indian Int. Conf. on Artificial Intelligence, IICAI-2007, Pune, India. find
  8. Bird, S., Klein, E., and Loper, E. (2009). Natural language processing with Python. O’Reilly Media. find
  9. Bottou, L. (1991). Stochastic gradient learning in neural networks. Proceedings of Neuro- Nımes , 91. find
  10. Carletta, J. (1996). Assessing agreement on classification tasks: the kappa statistic. Computational linguistics, 22(2):249–254. ↗
  11. Dakka, W. and Cucerzan, S. (2008). Augmenting Wikipedia with named entity tags. In Proceedings of the 3rd International Joint Conference on Natural Language Processing , pages 545–552, Hyderabad, India. Asian Federation of Natural Language Processing. find
  12. Diab, M. (2009). Second generation AMIRA tools for Arabic processing: Fast and robust tokenization, POS tagging, and base phrase chunking. In 2nd International Conference on Arabic Language Resources and Tools . find
  13. El-Halees, A. (2007). Arabic text classification using maximum entropy. The Islamic University Journal (Series of Natural Studies and Engineering) Vol , 15:157–167. find
  14. Elsebai, A., Meziane, F., and Belkredim, F. Z. (2009). A rule based persons names Arabic extraction system. Communications of the IBIMA, 11(6):53–59. find
  15. Farber, B., Freitag, D., Habash, N., and Rambow, O. (2008). Improving NER in Arabic using a morphological tagger. In Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), pages 2509–2514, Marrakech, Morocco. European Language Resources Association (ELRA).
  16. Habash, N., Soudi, A., and Buckwalter, T. (2007) On Arabic Transliteration. In A. van den Bosch and A. Soudi, editors, Arabic Computational Morphology: Knowledge-based and Empirical Methods. Springer. ↗
  17. Maloney, J. and Niv, M. (1998). Tagarab, a fast, accurate Arabic name recognizer using high-precision morphological analysis. In Proceedings of the Workshop on Computational Approaches to Semitic Languages , pages 8–15, Montreal, Quebec, Canada. Association for Computational Linguistics.
  18. Mohit, B., Schneider, N., Bhowmick, R., Oflazer, K., and Smith, N. (2012). Recall-oriented learning of named entities in Arabic Wikipedia. In Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2012) , pages 162– 173. Citeseer.
  19. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M. and Duchesnay, E. (2011) Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research, 12:2825-2830. ↗
  20. Saleh, I., Darwish, K., and Fahmy, A. (2010). Classifying Wikipedia articles into NE’s using SVM’s with threshold adjustment. In Proceedings of the 2010 Named Entities Workshop , pages 85–92, Uppsala, Sweden. Association for Computational Linguistics. find
  21. Shaalan, K. and Raza, H. (2007). Person name entity recognition for Arabic. In Proceedings of the 2007 Workshop on Computational Approaches to Semitic Languages: Common Issues and Resources, pages 17–24, Prague, Czech Republic. Association for Computational Linguistics.
  22. Shaalan, K. and Raza, H. (2009). NERA: Named entity recognition for Arabic. Journal of the American Society for Information Science and Technology, 60:1652–1663. ↗
  23. Syiam, M., Fayed, Z., and Habib, M. (2006). An intelligent system for Arabic text categorization. International Journal of Intelligent Computing and Information Sciences , 6(1):1–19. find
  24. Taghva, K., Elkhoury, R., and Coombs, J. (2005). Arabic stemming without a root dictionary. In Information Technology: Coding and Computing, 2005. ITCC 2005. International Conference on, volume 1, pages 152–157. IEEE. find
  25. Traboulsi, H. (2009). Arabic named entity extraction: A local grammar-based approach. In Proceedings of the 2009 International Multiconference on Computer Science and Information Technology (IMCSIT 2009), pages 139–143, Mragowo, Polland. find
  26. Yang, Y. and Pedersen, J. (1997). A comparative study on feature selection in text categorization. In Machine Learning-International Workshop Then Conference , pages 412–420. Morgan Kaufmann Publishers, INC. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alotaibi-2012-mapping-arabic-wikipedia-into-named/