Elamur Research⌕
☰
home / papers / 2013 / Using Arabic Wordnet for semantic indexation in information retrieval system

Using Arabic Wordnet for semantic indexation in information retrieval system

Paper · 2013 · Mohammed Alaeddine Abderrahim, Mohammed El Amine Abderrahim and Mohammed Amine Chikh

Summary

The paper asks whether Arabic information retrieval improves when documents and queries are indexed by meanings rather than surface keywords. Arabic WordNet serves as the sense inventory. After stop-word removal, the system matches terms against Arabic WordNet entries. A word that is not found is reduced to its root and looked up again, and words that are still unknown stay in the index as plain terms. For an ambiguous word, the system picks the synset most strongly connected, by semantic distance, to the senses of the other concepts in the same document. Queries go through the same procedure and are expanded with synonyms. The test collection has over 22,000 Arabic documents across several domains, searched with 70 keyword queries. Four setups are compared: plain keywords, semantic indexing of both sides, queries only, and documents only. Indexing both sides wins clearly on recall and on every precision cutoff. Indexing only one side can do worse than the keyword baseline, because the two vocabularies stop matching. The authors present this as the first concept-based indexing experiment for Arabic.

Key points

  • Corpus: about 22,000 Arabic documents (~180 MB, ~17M tokens, 612,650 distinct word forms) in domains such as health, sport, politics, science and religion, with 70 keyword queries.
  • Resource: Arabic WordNet, built on the Princeton WordNet / EuroWordNet design, with 11,269 synsets and 23,481 words.
  • Four setups:
    • R0: keyword baseline.
    • R1: semantic indexing of documents and queries.
    • R2: semantic indexing of queries only.
    • R3: semantic indexing of documents only.
  • Recall: R1 retrieved more documents and more relevant documents than R0 on all 70 queries. R3 retrieved fewer documents than R0 on half of the queries.
  • Precision: R1 beat R0 on every measure. The averaged precision score rose from about 0.40 to about 0.61, and top-rank precision from about 0.58 to 0.73. R2 and R3 landed between R0 and R1.

Interesting ideas

  • I1. Fallback to the root before giving up. When a surface form is missing from the lexicon, the system strips it to its base or root and retries the lookup, keeping the raw word only as a last resort. This raises lexicon coverage cheaply.
  • I2. Sense choice by coherence with the rest of the document. The chosen sense is the one most connected, by semantic distance, to the candidate senses of the other words in the same text. This collective, context-wide view replaces per-word guessing.
  • I3. Normalize both sides with the same pipeline. The paper's strongest finding is that replacing words with concepts on only one side breaks matching. A document indexed under one synonym is never found by a query that still uses another.
  • I4. Ablation by where the semantics is applied. Running query-only, document-only and both-sides variants against a baseline shows exactly which placement of a component helps or hurts.
  • I5. Count per-query wins and losses, not just averages. Besides mean precision, the authors report the share of queries that gained, stayed equal or lost results. This reveals regressions that averages hide.

For Mova

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

  • I1 (tokens): when a lemma has no WordNet/VerbNet entry, try a morphological back-off chain (derivational base, then stem), and log which step produced the match so the lookup stays explainable.
  • I2 (world): pick WordNet senses for entities in the text world by maximizing ontology links to the other entities' senses, and store the winning link as the justification.
  • I3 (answer): canonicalize questions and story facts through the identical lemma→sense mapping before matching answers, so a synonym in a fable question still hits the stored fact.
  • I4 (learn): evaluate a new knowledge layer such as sense mapping by enabling it on the input side, the knowledge side and both, treating the one-sided runs as controls that expose mismatch bugs.
  • I5 (learn): report per-item improved/unchanged/worse counts for every change on test suites, so a gain in the mean cannot conceal a group of broken questions.

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 Abou Bekr BelkaïdTlemcen · Algeria2 works · 21 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 3,066 worksInformation retrieval & search · 552 works

Cite

Cite as Mohammed Alaeddine Abderrahim, Mohammed El Amine Abderrahim and Mohammed Amine Chikh. 2013. Using Arabic Wordnet for semantic indexation in information retrieval system. arXiv:1306.2499 · IJCSI International Journal of Computer Science Issues, Vol. 10, Issue 1, No 2, January 2013.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1306.2499 · arXiv 1306.2499

@misc{abderrahim2013using,
  title = {{Using Arabic Wordnet for semantic indexation in information retrieval system}},
  author = {Mohammed Alaeddine Abderrahim, Mohammed {El Amine Abderrahim} and Mohammed Amine Chikh},
  year = {2013},
  doi = {10.48550/arxiv.1306.2499},
  eprint = {1306.2499},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1306.2499}
}

References

7 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. R. Mihalcea, D.I. Moldovan, Semantic Indexing using WordNet Senses, in Proceedings of ACL Workshop on IR & NLP; Hong Kong, 2000, pp 35-45. find
  2. M. Sanderson, Word Sense Disambiguation and Information Retrieval, PhD Thesis, Technical Report (TR-1997-7) of the Department of Computing Science at the University of Glasgow, Glasgow G12 8QQ, UK.10(1);1997. find
  3. M Baziz: Indexation conceptuelle guidée par ontologie pour la recherche d’information. Thèse de doctorat, université Paul Sabatier; 2005. find
  4. S. Elkateb, W. Black, H. Rodríguez, M. Alkhalifa, P. Vossen, A. Pease, C. Fellbaum: Building a WordNet for Arabic, in Proceedings of The fifth international conference on Language Resources and Evaluation; Genoa-Italy, 2006, pp 29-34.
  5. S. Elkateb, W. Black, P. Vossen, D. Farwell, A. Pease, C. Fellbaum: Arabic WordNet and the Challenges of Arabic. The Challenge of Arabic for NLP/MT. International conference at the British Computer Society, London, 2006; pp 15-24.
  6. W. Black, S. Elkateb, H. Rodriguez, M. Alkhalifa, P. Vossen, A. Pease, C. Fellbaum: Introducing the Arabic WordNet project, in Proceedings of the 3rd Global Wordnet Conference, Jeju Island, Korea, 2006, pp 295-299. find
  7. W. Black, Sabri, El-Kateb, A Prototype English-Arabic Dictionary Based on WordNet, proceedings of the Second International WordNet Conference, Brno, Czech Republic, 2004, pp 67-74. Mohammed Alaeddine Abderrahim is a research teacher at the University of Tiaret, Algeria. His research interests are natural language processing, information retrieval, information extraction, data mining and ontology. Med Alaeddine has a Magister in computer science from the University of Tlemcen, Algeria. He is a Doctorate candidate and a member of the Laboratory of Arabic Natural Language Processing in the University of Tlemcen. Mohammed El Amine Abderrahim is a research teacher at the University of Tlemcen, Algeria. His research interests are natural language processing, information retrieval, information extraction, databases and data mining. Med El Amine has a Magister in computer science from UST Oran, Algeria, and a Doctorate in computer science from the University of Tlemcen, Algeria. He is a member of the Laboratory of Arabic Natural Language Processing, university of Tlemcen. Mohammed Amine CHIKH received his Ph.D from the University of Tlemcen, Algeria and is currently a professor at the University of Tlemcen, Faculty of Technologie. His research interests include: knowledge engineering, knowledge discovery, data mining and medical semantic Web. Prof. Med Amine is a project manager in the Laboratory of Génie BioMédical (GBM), university of Tlemcen. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/abderrahim-2013-using-arabic-wordnet-semantic-indexation/