Elamur Research⌕
☰
home / papers / 2014 / Unsupervised Approach for Automatic Keyword Extraction from Arabic Documents

Unsupervised Approach for Automatic Keyword Extraction from Arabic Documents

Paper · 2014 · Atwi Awajan, Arafat

Summary

The paper tackles keyword extraction for Arabic. Few Arabic documents come with author keywords, and supervised extractors need annotated, domain-specific training data. Awajan proposes an unsupervised, single-document, domain-independent pipeline in two stages. First, text is split into sentences, clitics are separated (except the definite article), and tokens are POS-tagged with the Stanford Arabic parser. Candidate N-grams of up to three words are kept only if they consist of nouns and adjectives. Second, a morphological analyzer built on the Alkhalil Morpho-Syntactic System maps each word to a base form: the root (with its pattern) for derived words and the stem for non-derived ones. Candidates sharing base forms are merged, their counts are summed, and the top-weighted groups become keywords, each shown as its most frequent surface variant. On 70 journal articles from mixed domains, the method reaches an average precision of 0.51. The author argues this is respectable because only about 73% of the human keywords appear in the texts at all. The work shows that deep morphological grouping helps counter the data sparseness of a root-and-pattern language.

Key points

  • The evaluation corpus has 70 articles and abstracts drawn from six journals, split by length into three groups:
    • 22 documents averaging 6523 words;
    • 28 documents averaging 3238 words;
    • 20 documents averaging 212 words.
    • The overall average is 3406 words per document.
  • Gold keyword lists range from 2 to 14 per document (mean 5.14). The system outputs the same number as the gold list, so P, R and F coincide.
  • Precision is 0.56, 0.54 and 0.41 for the three groups, and 0.51 overall. Longer documents give better results.
  • Only about 73% of the manually assigned keywords occur in the document text, which caps achievable scores.
  • Tools: POS tagging relies on Green and Manning's (2010) Stanford parser for Arabic, and the morphological normalizer is built on the Alkhalil Morpho-Syntactic System. Related systems discussed include KP-Miner and El-Shishtawy & Al-Sammak's supervised extractor.

Interesting ideas

  • I1. Collapsing derived words to a shared root, while non-derived words fall back to their stem, lets morphologically different mentions of one concept pool their counts and stand out.
  • I2. Clitic splitting is applied repeatedly until the token no longer changes, which handles stacked attachments without enumerating every combination.
  • I3. The author measures how many gold keywords actually occur in the source text, turning a raw score into one judged against a reachable ceiling.
  • I4. Fixing the output size to the gold size makes precision, recall and F identical, giving one clean number per document.
  • I5. After clustering on normalized forms, each cluster is displayed by its most frequent original surface string, which keeps the output readable.

For Mova

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

  • I1 (world): Entity and concept salience in the text's world could be accumulated over derivational families (e.g. "decide/decision/decisive" via WordNet derivation links) rather than over exact lemmas.
  • I2 (tokens): The English tokenizer could run contraction and affix splitting to a fixpoint, so stacked forms like "y'all'd've" decompose cleanly.
  • I3 (learn): Evaluations of QA or extraction could report an oracle ceiling, such as the share of gold answers that are recoverable from the text, beside each accuracy figure.
  • I4 (learn): Ranked outputs such as salient entities or candidate frames could be evaluated at k equal to the gold count, avoiding threshold tuning in comparisons.
  • I5 (back): When generating text from the meaning graph, an entity or concept could be verbalized with its most frequent surface mention from the source, rather than a canonical lemma.

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

🎓Princess Sumaya University for TechnologyAmman · Jordan7 works · 63 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Atwi Awajan, Arafat. 2014. Unsupervised Approach for Automatic Keyword Extraction from Arabic Documents. Proceedings of the 26th Conference on Computational Linguistics and Speech Processing (ROCLING 2014).

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

@inproceedings{arafat2014unsupervised,
  title = {{Unsupervised Approach for Automatic Keyword Extraction from Arabic Documents}},
  author = {Atwi Awajan, Arafat},
  booktitle = {Proceedings of the 26th Conference on Computational Linguistics and Speech Processing (ROCLING 2014)},
  year = {2014},
  url = {https://aclanthology.org/O14-1018/}
}

References

20 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Al-Sughaier, I., Al-Kharashi, I. (2004). Arabic morphological analysis techniques: A comprehensive survey. Journal of The American Society for Information Science and Technology (JASIST), 55(3), 189-213. Average of words per article 6523 3238 212 3406 Precision 0.56 0.54 0.41 0.51 find
  2. Awajan, A. (2007). Arabic Text Preprocessing for the Natural Language Processing Applications. Arab Gulf Journal of Scientific Research, 25(4), 179-189. find
  3. Benajiba, Y., Diab, M., Rosso, P. (2009). Arabic Named Entity Recognition: A Feature-Driven Study. IEEE Transactions on Audio, Speech, and Language Processing, 17(5), 926-934. find
  4. Boudlal, A., Lakhouaja, A., Mazroui, A., Meziane, A., Ould Abdallahi, M., Shoul, M. (2010). Alkhalil Morpho Sys: A Morphosyntactic analysis system for Arabic texts, International Arab Conference on Information Technology (ACIT). Riyadh, Saudi Arabia. find
  5. Diab, M., Hacioglu, K., JURAFSKY, D. (2007). Automatic Processing of Modern Standard Arabic Text. Chapter in Arabic Computational Morphology. Springer Ed. 159-179. find
  6. El-Beltagy S., & Rafea A. (2008). KP-Miner: A keyphrase extraction system for English and Arabic documents, Information Systems. 34(1), 132-144. find
  7. El-Shishtawy, T., & Al-Sammak, A. (2009). Arabic Keyphrase Extraction using Linguistic knowledge and Machine Learning Techniques, In Proceedings of the Second International Conference on Arabic Language Resources and Tools, The MEDAR Consortium, Cairo, Egypt. find
  8. Frank, E., Paynter, G.W., Witten, I.H., Gutwin, C., Nevill-Manning, C.G. (1999). Domain-Specific Keyphrase Extraction. Proceedings of the International Joint Conference on Artificial Intelligence, Stockholm, Sweden, 668–673. find
  9. Green, S., and Manning, C. D. (2010). Better Arabic Parsing: Baselines, Evaluations, and Analysis. In COLING- Beijing. 394–402.
  10. Habash, N. (2010). Introduction to Arabic Natural Language Processing. Morgan & Claypool Publishers, USA. find
  11. Hasan, K.S., NG, V. (2010). Conundrums in unsupervised Keyphrase Extraction: Making Sense of the State-of-the-Art. COLING 2010, 365-373. find
  12. Hulth, A. (2003). Improved automatic keyword extraction given more linguistic knowledge. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, Sapporo, Japan, Hulth, A. (2004). Combining Machine Learning and Natural Language Processing for Automatic Keyword Extraction. Doctoral dissertation. Department of Computer and Systems Sciences, Stockholm University.
  13. Liu, Z., Li, P., Zheng, Y., Sun, M. (2009). Clustering to Find Exemplar Terms for Keyphrase Extraction. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, Singapore. 257–266. find
  14. Manning, C. D., Raghavan, P., Schütze, H. (2009). Introduction to Information Retrieval. Cambridge University Press. find
  15. Matsuo, Y., Ishizuka, M. (2004). Keyword Extraction from a Single Document using Word Co-occurrence Statistical Information. International Journal on Artificial Intelligence Tools, 13(1), 157-169 find
  16. Mihalcea, R., Tarau, P. (2004). TextRank: Brining order into texts. In Proceedings of EMNLP 2004, Association for Computational Linguistics, Barcelona, Spain. 404-411. find
  17. Pudota, N., Dattolo, A., Baruzzo, A., Tasso, C. (2010). A New Domain Independent Keyphrase Extraction System. Digital Libraries: Communications in Computer and Information Science, 91, 67-78. find
  18. Rose, S., Engel, D., Cramer, N., Cowley, W. (2010). Automatic keyword extraction from individual documents. Text Mining: Applications and Theory edited by Michael W. Berry and Jacob Kogan, John Wiley & Sons, Ltd. 3-20 find
  19. Turney, P. D. (2000). Learning Algorithm for Keyphrase Extraction. Information Retrieval, 2(4), 303-336. find
  20. Turney, P. D., & Pantel, P. (2010). From Frequency to Meaning: Vector Space Models of Semantics. Journal of Artificial Intelligence Research, 37, 141-188.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/awajan-2014-unsupervised-approach-automatic-keyword-extraction/