---
title: "Arabic Language Text Classification Using Dependency"
description: "The paper tests whether syntactic structure can replace frequency statistics when choosing the words… 12 references, cited by 1 paper here."
authors: ["Yannis Haralambous", "Yassir Elidrissi", "Philippe Lenca"]
year: 2014
venue: "arXiv:1410.4863"
source: "https://arxiv.org/abs/1410.4863"
doi: "10.48550/arxiv.1410.4863"
arxiv: "1410.4863"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/haralambous-2014-arabic-language-text-classification-using/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/haralambous-2014-arabic-language-text-classification-using/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2014](https://md.research.elamur.ai/free/en/papers/2014/page/3/) / Arabic Language Text Classification Using Dependency Syntax-Based Feature Selection

# Arabic Language Text Classification Using Dependency Syntax-Based Feature Selection

**Authors:** [Yannis Haralambous](https://md.research.elamur.ai/free/en/authors/yannis-haralambous/), Yassir Elidrissi and Philippe Lenca · **Year:** 2014 · **Venue:** arXiv:1410.4863 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/1410.4863) · [PDF](https://arxiv.org/pdf/1410.4863) · [DOI 10.48550/arxiv.1410.4863](https://doi.org/10.48550/arxiv.1410.4863) · [arXiv 1410.4863](https://arxiv.org/abs/1410.4863) · [OpenAlex](https://openalex.org/W2293726685)

**Citations:** 19 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/haralambous-2014-arabic-language-text-classification-using/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper tests whether syntactic structure can replace frequency statistics when choosing the words that represent an Arabic text for topic classification. The authors build a balanced set of 6,000 news documents across six topics from the Kalimat corpus, favoring documents with short, simple sentences. Text is prepared in two ways. One is clitic separation with the Stanford Word Segmenter, which amounts to very light stemming. The other replaces words with roots supplied by the Alkhalil analyzer. In both cases each unit is paired with a coarse POS tag. Sentences are parsed with the Stanford Parser, and features are chosen by their position in the dependency tree, either the head alone or the head plus nearby nouns. This is compared with tfidf-based selection. Class association rules and SVMs are trained per sentence, and each document's label comes from a confidence-weighted vote over its sentences. Light stemming outperforms roots. Small dependency-selected feature sets suit association rules, while SVMs do best with large, morphologically chosen sets. The work shows early on that a parse tree can serve as a compact, interpretable feature filter.

## Key points
- Data: 6,000 documents (1,000 per class) taken from the 20,291 in Kalimat. The six categories are Economy, Sports, Religion, Culture, International news and Local news (Oman). Documents were ranked by a score ρ that combines sentence-boundary density, log word count and inverse connective density. Sentence segmentation was checked by hand, and poetry, dialect and corrupted texts were removed.
- Features are a stem or root plus a simplified tag: n, np, v or x. Roots carry a special marker so they cannot collide with an identical surface word. Proper nouns are never rootified.
- The corpus contains 3,689 distinct roots. Of these, 1,441 occur in all six classes and 658 occur in only one class.
- Selection strategies range from the dependency head alone to the head plus nouns at tree distance 1, 2 or 3. The latter is nearly as discriminative as keeping every noun in the sentence.
- All computation, including tfidf, is done at the sentence level so the comparison is fair. The document class follows from a confidence-weighted majority vote, which is why scores are a bit below whole-document approaches.

## Interesting ideas
- **Tree-distance pruning:** a sentence's topical content can be captured by its syntactic head plus the nouns a few edges away, with little loss compared to all nouns.
- **Sentence-first, then vote:** predicting per sentence and combining by confidence-weighted voting keeps every document-level decision traceable to specific sentences.
- **Choosing parser-friendly texts by a simple score:** ranking documents by sentence shortness and low connective density yields input on which parsing is more reliable.
- **Namespaced feature keys:** pairing each unit with a coarse POS tag and marking roots separately prevents homographs and different abstraction levels from merging into one feature.
- **Match the learner to feature-set size:** rule-based classifiers thrive on small syntactically filtered feature sets, while SVMs need large ones. Compact structural features therefore pair naturally with interpretable rules.

## For Mova

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

- **I1** (frames): a head-plus-close-nominals "gist" could serve as a cheap, traceable key for choosing candidate frames or tagging a sentence's topic before full role assignment.
- **I2** (answer): readers answering about a whole fable could pool per-sentence evidence with weights and show which sentences drove the answer.
- **I3** (learn): a similar score could order silver-data candidates so that witness parsers first agree on easy texts, giving a curriculum with cleaner early training data.
- **I4** (tokens): lemma keys could carry a POS and level tag (lemma vs. derivational base), so downstream lookups never confuse, for example, a noun and verb sharing a spelling.
- **I5** (learn): self-induced repair rules could be mined over small, tree-filtered feature sets rather than full context windows, keeping the rules few and readable.

*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

Also on the byline: [Télécom Bretagne](https://ror.org/016p3g423).

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Yannis Haralambous, Yassir Elidrissi and Philippe Lenca. 2014. Arabic Language Text Classification Using Dependency Syntax-Based Feature Selection. arXiv:1410.4863.

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

```bibtex
@misc{haralambous2014arabic,
  title = {{Arabic Language Text Classification Using Dependency Syntax-Based Feature Selection}},
  author = {Yannis Haralambous, Yassir Elidrissi and Philippe Lenca},
  year = {2014},
  doi = {10.48550/arxiv.1410.4863},
  eprint = {1410.4863},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1410.4863}
}
```

- [← previous An Arabizi-English social media statistical machine translation system](https://md.research.elamur.ai/free/en/papers/may-2014-arabizi-english-social-media-statistical/)
- [next → Arabic Text Categorization Algorithm using Vector Evaluation Method](https://md.research.elamur.ai/free/en/papers/odeh-2015-arabic-text-categorization-algorithm-using/)
- [↑ back to papers of 2014](https://md.research.elamur.ai/free/en/papers/2014/page/3/)

## 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. Rakesh Agrawalh, Tomasz Imieliński, and Arun Swami. Mining association rules between sets of items in large databases. In Proceedings of the ACM SIGMOD International Conference on Management of Data 1993, Washington DC, USA, pages 207–216, 1993. [find](https://scholar.google.com/scholar?q=Rakesh+Agrawalh%2C+Tomasz+Imielin%CC%81ski%2C+and+Arun+Swami%2E+Mining+association+rules+between+sets+of+items+in+large+databases%2E+In+Proceedings+of+the+ACM+SIGMOD+International+Conference+on+Management+of+Data+)
2. Ahmed H. Al-Harbi, Abdulrahman Almuhareb, Abdulmohsen Al-Thubaity, Mohammad S. Khorsheed, and Abdullah Al-Rajeh. Automatic Arabic text classification. In Proceedings of Journées internationales d’analyse statistique des données textuelles 2008, Lyon, France, pages 77–83, 2008. [find](https://scholar.google.com/scholar?q=Ahmed+H%2E+Al%2DHarbi%2C+Abdulrahman+Almuhareb%2C+Abdulmohsen+Al%2DThubaity%2C+Mohammad+S%2E+Khorsheed%2C+and+Abdullah+Al%2DRajeh%2E+Automatic+Arabic+text+classification%2E+In+Proceedings+of+Journe%CC%81es+internationales+d%E2%80%99ana)
3. Ibrahim A. Al-Kharashi and Martha W. Evens. Comparing words, stems, and roots as index terms in an Arabic information retrieval system. Journal of the American Society for Information Science, 45(8):548–560, 1994. [↗](https://doi.org/10.1002/(sici)1097-4571(199409)45:8<548::aid-asi3>3.0.co;2-x)
4. Qasem Al-Radaideh, Emad Al-Shawakfa, Abdullah S. Ghareb, and Hani Abu-Salem. An approach for Arabic text categorization using association rule mining. International Journal of Computer Processing of Languages, 23(1):81–106, 2011. [↗](https://doi.org/10.1142/s179384061100222x)
5. Bassam Al-Salemi and Mohammad Juzaiddin Ab Aziz. Statistical Bayesian learning for automatic Arabic text categorization. Journal of Computer Science, 7(1):39–45, 2011. [↗](https://doi.org/10.3844/jcssp.2011.39.45)
6. Riyad Al-Shalabi and Rasha Obeidat. Improving kNN Arabic text classification with n-grams based document indexing. In Proceedings of INFOS 2008, Cairo, Egypt, pages NLP–108–112, 2008. [find](https://scholar.google.com/scholar?q=Riyad+Al%2DShalabi+and+Rasha+Obeidat%2E+Improving+kNN+Arabic+text+classification+with+n%2Dgrams+based+document+indexing%2E+In+Proceedings+of+INFOS+2008%2C+Cairo%2C+Egypt%2C+pages+NLP%E2%80%93108%E2%80%93112%2C+2008%2E)
7. Eiman Tamah Al-Shammari. Improving Arabic document categorization: Introducing local stem. In Proceedings of ISDA 2010, Cairo, Egypt, pages 385–390, 2010. [find](https://scholar.google.com/scholar?q=Eiman+Tamah+Al%2DShammari%2E+Improving+Arabic+document+categorization%3A+Introducing+local+stem%2E+In+Proceedings+of+ISDA+2010%2C+Cairo%2C+Egypt%2C+pages+385%E2%80%93390%2C+2010%2E)
8. Abdulmohsen Al-Thubaity, Norah Abanumay, Sara Al-Jerayyed, Aljo-harah Alrukban, and Zarah Mannaa. The effect of combining different feature selection methods on Arabic text classification. In Proceedings of ACIS/SPND 2013, Honolulu, Hawaii, USA, pages 211–216, 2013. [find](https://scholar.google.com/scholar?q=Abdulmohsen+Al%2DThubaity%2C+Norah+Abanumay%2C+Sara+Al%2DJerayyed%2C+Aljo%2Dharah+Alrukban%2C+and+Zarah+Mannaa%2E+The+effect+of+combining+different+feature+selection+methods+on+Arabic+text+classification%2E+In+Proceedi)
9. Aya Al-Zoghby, Ahmed Sharaf Eldin, Nabil A. Ismail, and Taher Hamza. Mining Arabic text using soft-matching association rules. Proceedings of ICCES 2007, Miami, USA, pages 421–426, 2007. [find](https://scholar.google.com/scholar?q=Aya+Al%2DZoghby%2C+Ahmed+Sharaf+Eldin%2C+Nabil+A%2E+Ismail%2C+and+Taher+Hamza%2E+Mining+Arabic+text+using+soft%2Dmatching+association+rules%2E+Proceedings+of+ICCES+2007%2C+Miami%2C+USA%2C+pages+421%E2%80%93426%2C+2007%2E)
10. Mohammed Aljlayl and Ophir Frieder. On Arabic search: Improving the retrieval effectiveness via a light stemming approach. In Proceedings of CIKM 2002, McLean, VA, USA, pages 340–347, 2002. [find](https://scholar.google.com/scholar?q=Mohammed+Aljlayl+and+Ophir+Frieder%2E+On+Arabic+search%3A+Improving+the+retrieval+effectiveness+via+a+light+stemming+approach%2E+In+Proceedings+of+CIKM+2002%2C+McLean%2C+VA%2C+USA%2C+pages+340%E2%80%93347%2C+2002%2E)
11. Jaber Alwedyan, Wa’el Musa Hadi, Ma’an Salam, and Hussein Y. Mansour. Categorize Arabic data sets using multi-class classification based on association rule approach. In Proceedings of ISWSA 2011, Amman, Jordan, pages 18:1–18:8, 2011. [find](https://scholar.google.com/scholar?q=Jaber+Alwedyan%2C+Wa%E2%80%99el+Musa+Hadi%2C+Ma%E2%80%99an+Salam%2C+and+Hussein+Y%2E+Mansour%2E+Categorize+Arabic+data+sets+using+multi%2Dclass+classification+based+on+association+rule+approach%2E+In+Proceedings+of+ISWSA+2011%2C+Amm)
12. Mohammed Azara, Tamer Fatayer, and Alaa El-Halees. Arabic text classification using Learning Vector Quantization. In Proceedings of INFOS 2012, Cairo, Egypt, pages NLP–39–43, 2012. IV′ [find](https://scholar.google.com/scholar?q=Mohammed+Azara%2C+Tamer+Fatayer%2C+and+Alaa+El%2DHalees%2E+Arabic+text+classification+using+Learning+Vector+Quantization%2E+In+Proceedings+of+INFOS+2012%2C+Cairo%2C+Egypt%2C+pages+NLP%E2%80%9339%E2%80%9343%2C+2012%2E+IV%E2%80%B2)

*Updated 1 November 2026*