---
title: "TPT: An Empirical Term Selection for Arabic Text Categorization — Abbas 2021"
description: "The authors test the Transition Point (TP) technique, which comes from Zipf's and Booth's work on… 90.75% recall. 18 references, cited by 1 paper here."
authors: ["Mourad Abbas", "Mohamed Lichouri"]
year: 2021
venue: "Proceedings of the 4th International Conference on Natural Language and Speech Processing (ICNLSP 2021)"
source: "https://aclanthology.org/2021.icnlsp-1.26/"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/abbas-2021-tpt-empirical-term-selection-arabic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/abbas-2021-tpt-empirical-term-selection-arabic/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2021](https://md.research.elamur.ai/free/en/papers/2021/page/5/) / TPT: An Empirical Term Selection for Arabic Text Categorization

# TPT: An Empirical Term Selection for Arabic Text Categorization

**Authors:** [Mourad Abbas](https://md.research.elamur.ai/free/en/authors/mourad-abbas/) and [Mohamed Lichouri](https://md.research.elamur.ai/free/en/authors/mohamed-lichouri/) · **Year:** 2021 · **Venue:** Proceedings of the 4th International Conference on Natural Language and Speech Processing (ICNLSP 2021) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2021.icnlsp-1.26/) · [PDF](https://aclanthology.org/2021.icnlsp-1.26.pdf)

**Citations:** 1 in our library, of the 140,884 papers whose bibliographies we read

## Summary
The paper asks whether a cheap, frequency-based rule for choosing index terms can match standard feature selection in Arabic text categorization. The authors test the Transition Point (TP) technique, which comes from Zipf's and Booth's work on word frequencies. TP computes a frequency value from the count of words that occur only once. It then keeps only the terms whose frequencies fall in a window around that value, on the idea that mid-frequency words carry the most content. TP is compared with Term Frequency (TF) and Document Frequency (DF) selection. All three feed the same TFIDF classifier, which assigns documents to categories by cosine similarity. Experiments use the Khaleej-2004 newspaper corpus, which has four categories. The three methods score almost the same: DF is marginally best and TP marginally last. However, TP needs a vocabulary of roughly 2,500 words, while TF and DF use about 40,000. This makes TP attractive when compact representations matter. The authors note that the threshold is still set empirically.

## Key points
- **Corpus.** Khaleej-2004 contains 5,690 articles from an Arabic online newspaper in four categories: Economy, International news, Local news and Sports. It has about 2,853,000 words. Removing punctuation and stop words shrinks it by 23.90%, to 2,172,000 words.
- **TP formula.** The TP value is (√(8·I₁+1) − 1)/2, where I₁ is the number of hapax words. Terms are kept if their frequency lies between (1−NTP)·TP and (1+NTP)·TP, with NTP in [0,1].
- **Results.** With NTP = 0.9, TP reaches 90.75% recall and 90.5% precision. TF gets 90.80% and 91.20%, and DF gets 91.75% and 91.50%.
- **Effect of NTP.** For all four categories, performance rises as NTP grows.
- **Why raw frequency misleads.** A category's most frequent words can still be uninformative. For example, "year" (\’ A m\) ranks near the top in every category, whereas "match" (\m b A r A t\) is concentrated in Sports.

## Interesting ideas
- The selection threshold comes from a closed-form statistic of the text itself (the hapax count), not from a tuned hyperparameter, so it adapts automatically to corpus size and language.
- Words are judged by how their relative frequency compares across classes, not by raw frequency, which exposes frequent but non-discriminative words like "year".
- Vocabulary size is reported alongside accuracy, showing that a 16× smaller lexicon costs only about one point of recall.
- Sweeping the window-width parameter and plotting per-category recall and precision reveals how each class responds, not just a single best score.
- The paper is motivated by the observation that mid-frequency words carry the most content, while very frequent and very rare ones are noise or function words.

## For Mova

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

- **I1** (learn): The same kind of self-derived cutoff could decide which mid-frequency words or constructions in silver data deserve LLM-teacher review, without hand-tuning a cutoff per corpus.
- **I2** (world): Comparing how often a noun or verb appears across genres such as fables, school tests and word problems could flag which entities and verb classes are genre markers, helping the world builder decide which knowledge to load first.
- **I3** (answer): Evaluation of the readers could report how many lexicon entries or frames each answer path actually relied on, rewarding compact, explainable coverage over sprawling resources.
- **I4** (tree): Per-relation LAS curves under a sweep, for example of beam size or the threshold for accepting silver data, would show which UD relations degrade first.
- **I5** (tokens): The en crate could focus manual lemma and morphology checks on the mid-frequency band of a corpus, where errors most affect content understanding.

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

Fields: [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/)

Topics: [🐪 Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Cite

Cite as: Mourad Abbas and Mohamed Lichouri. 2021. TPT: An Empirical Term Selection for Arabic Text Categorization. Proceedings of the 4th International Conference on Natural Language and Speech Processing (ICNLSP 2021).

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

```bibtex
@inproceedings{abbas2021tpt,
  title = {{TPT: An Empirical Term Selection for Arabic Text Categorization}},
  author = {Mourad Abbas and Mohamed Lichouri},
  booktitle = {Proceedings of the 4th International Conference on Natural Language and Speech Processing (ICNLSP 2021)},
  year = {2021},
  url = {https://aclanthology.org/2021.icnlsp-1.26/}
}
```

- [← previous Thematic role assignment in non-default verb classes: A cross-linguistic comparison of English and German](https://md.research.elamur.ai/free/en/papers/kyriaki-2021-thematic-role-assignment-non-default/)
- [next → Validation of Universal Dependencies by regeneration](https://md.research.elamur.ai/free/en/papers/lapalme-2021-validation-universal-dependencies-regeneration/)
- [↑ back to papers of 2021](https://md.research.elamur.ai/free/en/papers/2021/page/5/)

## References

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

1. M Abbas, K Smaili, and D Berkani. 2011. Evaluation of topic identification methods for arabic texts and their combination by using a corpus extracted from the omani newspaper alwatan. Arab Gulf Journal of Scientific Research, 29(3-4):183–191. [find](https://scholar.google.com/scholar?q=M+Abbas%2C+K+Smaili%2C+and+D+Berkani%2E+2011%2E+Evaluation+of+topic+identification+methods+for+arabic+texts+and+their+combination+by+using+a+corpus+extracted+from+the+omani+newspaper+alwatan%2E+Arab+Gulf+Journa)
2. Mourad Abbas and Kamel Smaili. 2005. Comparison of topic identification methods for arabic language. In Proceedings of International Conference on Recent Advances in Natural Language Processing, RANLP, pages 14–17. [find](https://scholar.google.com/scholar?q=Mourad+Abbas+and+Kamel+Smaili%2E+2005%2E+Comparison+of+topic+identification+methods+for+arabic+language%2E+In+Proceedings+of+International+Conference+on+Recent+Advances+in+Natural+Language+Processing%2C+RANLP)
3. Rubı Cabrera, David Pinto, H Jimenez, and D Vilarino. 2005. Una nueva ponderación para el modelo de espacio vectorial de recuperación de información. Research on Computing Science, 13:75–81. [find](https://scholar.google.com/scholar?q=Rub%C4%B1+Cabrera%2C+David+Pinto%2C+H+Jimenez%2C+and+D+Vilarino%2E+2005%2E+Una+nueva+ponderaci%C3%B3n+para+el+modelo+de+espacio+vectorial+de+recuperaci%C3%B3n+de+informaci%C3%B3n%2E+Research+on+Computing+Science%2C+13%3A75%E2%80%9381%2E)
4. RC Eldridge. 1911. Six Thousand Common English Words: Their Comparative Frequency and What Can Be Done with Them. Clement Press. [find](https://scholar.google.com/scholar?q=RC+Eldridge%2E+1911%2E+Six+Thousand+Common+English+Words%3A+Their+Comparative+Frequency+and+What+Can+Be+Done+with+Them%2E+Clement+Press%2E)
5. Héctor Jimenez, David Pinto, and Paolo Rosso. 2005. Selección de términos no supervisada para agru-pamiento de resúmenes. In proceedings of Workshop on Human Language, ENC05, pages 86–91. [find](https://scholar.google.com/scholar?q=H%C3%A9ctor+Jimenez%2C+David+Pinto%2C+and+Paolo+Rosso%2E+2005%2E+Selecci%C3%B3n+de+t%C3%A9rminos+no+supervisada+para+agru%2Dpamiento+de+res%C3%BAmenes%2E+In+proceedings+of+Workshop+on+Human+Language%2C+ENC05%2C+pages+86%E2%80%9391%2E)
6. Edgar Moyotl Hernández and Héctor Jiménez Salazar. 2004. An analysis on frequency of terms for text categorization. Procesamiento del lenguaje natural, nº 33 (septiembre 2004); pp. 141-146. [find](https://scholar.google.com/scholar?q=Edgar+Moyotl+Hern%C3%A1ndez+and+H%C3%A9ctor+Jim%C3%A9nez+Salazar%2E+2004%2E+An+analysis+on+frequency+of+terms+for+text+categorization%2E+Procesamiento+del+lenguaje+natural%2C+n%C2%BA+33+%28septiembre+2004%29%3B+pp%2E+141%2D146%2E)
7. Edgar Moyotl-Hernández and Héctor Jiménez-Salazar. 2005. Enhancement of dtp feature selection method for text categorization. In International Conference on Intelligent Text Processing and Computational Linguistics, pages 719–722. Springer. [find](https://scholar.google.com/scholar?q=Edgar+Moyotl%2DHern%C3%A1ndez+and+H%C3%A9ctor+Jim%C3%A9nez%2DSalazar%2E+2005%2E+Enhancement+of+dtp+feature+selection+method+for+text+categorization%2E+In+International+Conference+on+Intelligent+Text+Processing+and+Computation)
8. David Pinto, Héctor Jiménez-Salazar, and Paolo Rosso. 2006. Clustering abstracts of scientific texts using the transition point technique. In International Conference on Intelligent Text Processing and Computational Linguistics, pages 536–546. Springer. [find](https://scholar.google.com/scholar?q=David+Pinto%2C+H%C3%A9ctor+Jim%C3%A9nez%2DSalazar%2C+and+Paolo+Rosso%2E+2006%2E+Clustering+abstracts+of+scientific+texts+using+the+transition+point+technique%2E+In+International+Conference+on+Intelligent+Text+Processing+an)
9. Joseph Rocchio. 1971. Relevance feedback in information retrieval. The Smart retrieval system-experiments in automatic document processing, pages 313–323. [find](https://scholar.google.com/scholar?q=Joseph+Rocchio%2E+1971%2E+Relevance+feedback+in+information+retrieval%2E+The+Smart+retrieval+system%2Dexperiments+in+automatic+document+processing%2C+pages+313%E2%80%93323%2E)
10. Ronald Rosenfeld and Xuedong Huang. 1992. Improvements in stochastic language modeling. In Speech and Natural Language: Proceedings of a Workshop Held at Harriman, New York, February 23-26, 1992. [find](https://scholar.google.com/scholar?q=Ronald+Rosenfeld+and+Xuedong+Huang%2E+1992%2E+Improvements+in+stochastic+language+modeling%2E+In+Speech+and+Natural+Language%3A+Proceedings+of+a+Workshop+Held+at+Harriman%2C+New+York%2C+February+23%2D26%2C+1992%2E)
11. Gerard Salton. 1991. Developments in automatic text retrieval. science, 253(5023):974–980. [find](https://scholar.google.com/scholar?q=Gerard+Salton%2E+1991%2E+Developments+in+automatic+text+retrieval%2E+science%2C+253%285023%29%3A974%E2%80%93980%2E)
12. Mireya Tovar, Maya Carrillo, David Pinto, and H Jimenez. 2005. Combining keyword identification techniques. Research on Computing Science, 14:157–162. [find](https://scholar.google.com/scholar?q=Mireya+Tovar%2C+Maya+Carrillo%2C+David+Pinto%2C+and+H+Jimenez%2E+2005%2E+Combining+keyword+identification+techniques%2E+Research+on+Computing+Science%2C+14%3A157%E2%80%93162%2E)
13. Mourad Abbas, Kamel Smaïli, and Daoud Berkani. 2010. Efficiency of tr-classifier versus tfidf. In 2010 First International Conference on Integrated Intelligent Computing, pages 233–237. IEEE. [find](https://scholar.google.com/scholar?q=Mourad+Abbas%2C+Kamel+Sma%C3%AFli%2C+and+Daoud+Berkani%2E+2010%2E+Efficiency+of+tr%2Dclassifier+versus+tfidf%2E+In+2010+First+International+Conference+on+Integrated+Intelligent+Computing%2C+pages+233%E2%80%93237%2E+IEEE%2E)
14. Yiming Yang and Jan O Pedersen. 1997. A comparative study on feature selection in text categorization. In Icml, volume 97, page 35. Nashville, TN, USA. [find](https://scholar.google.com/scholar?q=Yiming+Yang+and+Jan+O+Pedersen%2E+1997%2E+A+comparative+study+on+feature+selection+in+text+categorization%2E+In+Icml%2C+volume+97%2C+page+35%2E+Nashville%2C+TN%2C+USA%2E)
15. Andrew D Booth. 1967. A “law of occurrences for words of low frequency. Information and control, 10(4):386–393. [find](https://scholar.google.com/scholar?q=Andrew+D+Booth%2E+1967%2E+A+%E2%80%9Claw+of+occurrences+for+words+of+low+frequency%2E+Information+and+control%2C+10%284%29%3A386%E2%80%93393%2E)
16. George Kingsley Zipf. 2016. Human behavior and the principle of least effort: An introduction to human ecology. Ravenio Books. [find](https://scholar.google.com/scholar?q=George+Kingsley+Zipf%2E+2016%2E+Human+behavior+and+the+principle+of+least+effort%3A+An+introduction+to+human+ecology%2E+Ravenio+Books%2E)
17. Armelle Brun. 2003. Détection de thème et adaptation des modèles de langage pour la reconnaissance automatique de la parole. Ph.D. thesis, Université Henri Poincaré-Nancy 1. [find](https://scholar.google.com/scholar?q=Armelle+Brun%2E+2003%2E+D%C3%A9tection+de+th%C3%A8me+et+adaptation+des+mod%C3%A8les+de+langage+pour+la+reconnaissance+automatique+de+la+parole%2E+Ph%2ED%2E+thesis%2C+Universit%C3%A9+Henri+Poincar%C3%A9%2DNancy+1%2E)
18. Claudia Bueno, David Pinto, and Héctor Jiménez. 2005. El párrafo virtual en la generación de extractos. Research on Computing Science, 13:83–90. [find](https://scholar.google.com/scholar?q=Claudia+Bueno%2C+David+Pinto%2C+and+H%C3%A9ctor+Jim%C3%A9nez%2E+2005%2E+El+p%C3%A1rrafo+virtual+en+la+generaci%C3%B3n+de+extractos%2E+Research+on+Computing+Science%2C+13%3A83%E2%80%9390%2E)