---
title: "Authorship Verification for Arabic Short Texts Using Arabic"
description: "The authors build AraKB, a hand-made lexical knowledge base covering the 1,000 most… 0.77 accuracy. 18 references, cited by 1 paper here."
authors: ["Fatimah Alqahtani", "Helen Yannakoudakis"]
year: 2022
venue: "Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)"
source: "https://aclanthology.org/2022.wanlp-1.19/"
doi: "10.18653/v1/2022.wanlp-1.19"
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/alqahtani-2022-authorship-verification-arabic-short-texts/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/alqahtani-2022-authorship-verification-arabic-short-texts/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2022](https://md.research.elamur.ai/free/en/papers/2022/page/8/) / Authorship Verification for Arabic Short Texts Using Arabic Knowledge-Base Model (AraKB)

# Authorship Verification for Arabic Short Texts Using Arabic Knowledge-Base Model (AraKB)

**Authors:** Fatimah Alqahtani and [Helen Yannakoudakis](https://md.research.elamur.ai/free/en/authors/helen-yannakoudakis/) · **Year:** 2022 · **Venue:** [Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)](https://md.research.elamur.ai/free/en/venues/wanlp/) ([WANLP 2022](https://md.research.elamur.ai/free/en/venues/wanlp/2022/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2022.wanlp-1.19/) · [PDF](https://aclanthology.org/2022.wanlp-1.19.pdf) · [DOI 10.18653/v1/2022.wanlp-1.19](https://doi.org/10.18653/v1/2022.wanlp-1.19) · [OpenAlex](https://openalex.org/W4385573041)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/alqahtani-2022-authorship-verification-arabic-short-texts/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper addresses authorship verification (AV), which asks whether two texts share an author, for short Arabic social media posts. This setting has been studied little. Tweets are brief, and Arabic morphology, diglossia and dialect complicate matters further. AV also cannot rely on heavy normalization, because normalization erases the stylistic traces it depends on. The authors build AraKB, a hand-made lexical knowledge base covering the 1,000 most frequent items in their tweet corpus. Each entry receives graded scores between 0 and 1 on a set of meaning-related dimensions, plus part-of-speech indicator columns. A tweet becomes the average of its known items' vectors, and a word that follows a negator has its scores sign-inverted. These meaning vectors are added to stylometric features from the authors' earlier work and classified with Gradient Boosting. The result is a 2-point gain over stylometry alone, reaching 0.77 accuracy and F1. That matches a stylometry + TF-IDF setup while avoiding dependence on specific words. The work is preliminary but suggests that explicit, interpretable meaning profiles can carry authorial signal.

## Key points
- **Dataset:** reused from Alqahtani & Dohler (2022a). It has 100 Twitter users writing in a mix of Gulf dialect and MSA, with 375,428 tweets in total and 1,000–3,000 tweets per user.
- **AraKB contents:** 1,000 entries, made up of 895 words, 15 punctuation marks and 90 emojis.
  - Names, usernames and English words were left out.
  - Ordinary stop words were dropped, but all negators were retained, including MSA and dialectal forms.
- **Representation:**
  - Each feature column holds a value from 0 (opposite meaning) to 1 (exact meaning).
  - Part of speech is split into one column per tag, because categories cannot be averaged.
  - Tweets with no AraKB word are ignored.
- **Setup and results:** the setup matches the earlier study, with Gradient Boosting, 5-fold CV and the same preprocessing.
  - Stylometric features alone: 0.75 F1 and 0.75 accuracy.
  - Stylometric + TF-IDF: 0.77 F1 and 0.77 accuracy.
  - Stylometric + AraKB: 0.77 on F1, recall, precision and accuracy.
- **Overfitting argument:** the authors argue AraKB is not overfitted to the corpus, because the most frequent words are shared by most users and are therefore not author-specific. A linguist manually reviewed the table.

## Interesting ideas
- A manually scored table that places each frequent word on graded meaning dimensions turns a text into an averaged, human-readable "meaning profile" instead of an opaque embedding or word counts.
- Negation is handled by flipping the sign of the next word's feature scores before averaging, so "not honest" lowers the tweet's honesty instead of raising it.
- Emojis and punctuation are treated as full lexical items that carry meaning, not as noise. Stop-word filtering explicitly spares negators, including dialectal ones.
- The new features are tested under exactly the prior study's split, classifier, CV and preprocessing. They are also compared against a strong content baseline (TF-IDF), so the effect of the feature set is isolated.
- Coverage is limited on purpose to the most frequent shared vocabulary, which the authors argue keeps features from memorizing author-specific words.

## For Mova

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

- **I1** (world): A compiled lexicon of graded value dimensions keyed by lemma (honesty, positivity, and similar) could give every text an explainable attitude profile on top of the entity/state world.
- **I2** (frames): The same polarity inversion should be applied, but its scope should come from the UD `neg`/`advmod` attachment in the tree rather than from linear adjacency. That keeps the flip explainable and correct for longer scopes.
- **I3** (tokens): The tokenizer and lemma layer could give emojis and expressive punctuation lexical entries with sentiment-like attributes. Negator lists should be protected from any filtering step.
- **I4** (learn): When adding a new representation, rerun the frozen previous pipeline and a bag-of-words control side by side, so that any gain, or a mere tie as here, is attributable.
- **I5** (answer): Readers could prioritize hand-curated knowledge for the high-frequency core vocabulary of school texts and fables, where effort pays off across most inputs.

*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

- [King's College London](https://md.research.elamur.ai/free/en/universities/king-s-college-london/) — University, London, United Kingdom · 42 works · 1,617 citations

📚 Shelves (5): [🖥️ 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 · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,437 works

🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🕸️ Knowledge graphs & ontologies](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/) · 192 works

## Cite

Cite as: Fatimah Alqahtani and Helen Yannakoudakis. 2022. Authorship Verification for Arabic Short Texts Using Arabic Knowledge-Base Model (AraKB). Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP).

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

```bibtex
@inproceedings{alqahtani2022authorship,
  title = {{Authorship Verification for Arabic Short Texts Using Arabic Knowledge-Base Model (AraKB)}},
  author = {Fatimah Alqahtani and Helen Yannakoudakis},
  booktitle = {Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)},
  year = {2022},
  doi = {10.18653/v1/2022.wanlp-1.19},
  url = {https://doi.org/10.18653/v1/2022.wanlp-1.19}
}
```

- [← previous Assessment of Sales Negotiation Strategies with ISO 24617-2 Dialogue Act Annotations](https://md.research.elamur.ai/free/en/papers/stock-2022-assessment-sales-negotiation-strategies-iso/)
- [next → Automated detection of pronunciation errors in non-native English speech employing deep learning](https://md.research.elamur.ai/free/en/papers/korzekwa-2022-automated-detection-pronunciation-errors-non/)
- [↑ back to papers of 2022](https://md.research.elamur.ai/free/en/papers/2022/page/8/)

## References

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

1. Ahmed Abbasi and Hsinchun Chen. 2005. Applying authorship analysis to extremist-group Web forum messages. IEEE Intelligent Systems, 20(5), 67–75. https://doi.org/10.1109/MIS.2005.81 [↗](https://doi.org/10.1109/mis.2005.81)
2. Bouzoubaa, K., Baidouri, H., Loukili, T., & Yazidi, T. El. (2009). Arabic stop words: Towards a generalisation and standardisation. In Knowledge Management and Innovation in Advancing Economies: Analyses and Solutions -Proceedings of the 13th International Business Information Management Association Conference, IBIMA 2009, 3, 1844–1848. [find](https://scholar.google.com/scholar?q=Bouzoubaa%2C+K%2E%2C+Baidouri%2C+H%2E%2C+Loukili%2C+T%2E%2C+%26+Yazidi%2C+T%2E+El%2E+%282009%29%2E+Arabic+stop+words%3A+Towards+a+generalisation+and+standardisation%2E+In+Knowledge+Management+and+Innovation+in+Advancing+Economies%3A+Analy)
3. El-Said Badawi. 1996. Understanding Arabic: essays in contemporary Arabic linguistics in honor of El-Said Badawi. American Univ in Cairo Press. [find](https://scholar.google.com/scholar?q=El%2DSaid+Badawi%2E+1996%2E+Understanding+Arabic%3A+essays+in+contemporary+Arabic+linguistics+in+honor+of+El%2DSaid+Badawi%2E+American+Univ+in+Cairo+Press%2E)
4. Fatimah Alqahtani and Mischa Dohler. 2022a. Investigating Predictive Features for Authorship Verification of Arabic Tweets. IJCSNS, 22(6), 115. [find](https://scholar.google.com/scholar?q=Fatimah+Alqahtani+and+Mischa+Dohler%2E+2022a%2E+Investigating+Predictive+Features+for+Authorship+Verification+of+Arabic+Tweets%2E+IJCSNS%2C+22%286%29%2C+115%2E)
5. Fatimah Alqahtani and Mischa Dohler. 2022b. Survey of Authorship Identification Tasks on Arabic Texts. In ACM Transactions on Asian and Low-Resource Language Information Processing. [↗](https://doi.org/10.1145/3564156)
6. Hosein Azarbonyad, Mostafa Dehghani, Maarten Marx, and Jaap Kamps. 2015. Time-Aware Authorship Attribution for Short Text Streams. In Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages. 727-730. [find](https://scholar.google.com/scholar?q=Hosein+Azarbonyad%2C+Mostafa+Dehghani%2C+Maarten+Marx%2C+and+Jaap+Kamps%2E+2015%2E+Time%2DAware+Authorship+Attribution+for+Short+Text+Streams%2E+In+Proceedings+of+the+38th+International+ACM+SIGIR+Conference+on+Rese)
7. Hossam Ahmed. 2017. Dynamic Similarity Threshold in Authorship Verification: Evidence from Classical Arabic. In Procedia of The 3rd International Conference on Arabic Computational Linguistics, pages 145–152. https://doi.org/10.1016/j.procs.2017.10.103 [↗](https://doi.org/10.1016/j.procs.2017.10.103)
8. Hossam Ahmed. 2018. The Role of Linguistic Feature Categories in Authorship Verification. In Procedia of The 4th International Conference on Arabic Computational Linguistics, 142, pages 214–221. [find](https://scholar.google.com/scholar?q=Hossam+Ahmed%2E+2018%2E+The+Role+of+Linguistic+Feature+Categories+in+Authorship+Verification%2E+In+Procedia+of+The+4th+International+Conference+on+Arabic+Computational+Linguistics%2C+142%2C+pages+214%E2%80%93221%2E)
9. Hossam Ahmed. 2019a. [Distance-Based Authorship Verification Across Modern Standard Arabic Genres](https://md.research.elamur.ai/free/en/papers/ahmed-2019-distance-based-authorship-verification-across/). In of the 3rd workshop on arabic corpus linguistics, pages 89-96
10. Hossam Ahmed. 2019b. [Sample Size in Arabic Authorship Verification](https://md.research.elamur.ai/free/en/papers/ahmed-2019-sample-size-arabic-authorship-verification/). In Proceedings of the 3rd International Conference on Natural Language and Speech Processing, pages 84-91. Association for Computational Linguistics.
11. Jalaluddin Al-Suyuti. 1998. Al-Mazhar fi ’ulum al-lughat wa ’iwa’aha. Almaktabat aleasria. [find](https://scholar.google.com/scholar?q=Jalaluddin+Al%2DSuyuti%2E+1998%2E+Al%2DMazhar+fi+%E2%80%99ulum+al%2Dlughat+wa+%E2%80%99iwa%E2%80%99aha%2E+Almaktabat+aleasria%2E)
12. Karin C. Ryding. 2005. A Reference Grammar of Modern Standard Arabic. Cambridge university press. https://doi.org/https://doi.org/10.1017/CBO978 0511486975 [↗](https://doi.org/10.1017/cbo978)
13. Khaled Shaalan and Hafsa Raza. 2009. NERA: Named entity recognition for Arabic. In Journal of the American Society for Information Science and Technology, 60(8), pages 1652–1663. https://doi.org/10.1002/asi.21090 [↗](https://doi.org/10.1002/asi.21090)
14. Kim Luyckx and Walter Daelemans. 2011. The effect of author set size and data size in authorship attribution. 26(1), pages 35-55 https://doi.org/10.1093/llc/fqq013 [↗](https://doi.org/10.1093/llc/fqq013)
15. Leah S. Larkey and Margaret E. Connell. 2001. Arabic Information Retrieval at UMass in TREC-10. In The Tenth Text REtrieval Conference. [find](https://scholar.google.com/scholar?q=Leah+S%2E+Larkey+and+Margaret+E%2E+Connell%2E+2001%2E+Arabic+Information+Retrieval+at+UMass+in+TREC%2D10%2E+In+The+Tenth+Text+REtrieval+Conference%2E)
16. Mary L. McHugh. 2012. Interrater reliability : the kappa statistic. In Biochemica Medica, 22(3), pages 276–282. Retrieved from https://hrcak.srce.hr/89395 [↗](https://doi.org/10.11613/bm.2012.031)
17. Mohammad Al-Sarem, Walid Cherif, Ahmed Abdel Wahab, Abdel-Hamid Emara, and Mohamed Kissi. 2018. Combination of stylo-based features and frequency-based features for identifying the author of short Arabic text. In of the 12th International Conference on Intelligent Systems: Theories and Applications, pages 1-6. https://doi.org/10.1145/3289402.3289500 [↗](https://doi.org/10.1145/3289402.3289500)
18. Sushil Kumar and Mousmi A. Chaurasia. 2012. Assessment on Stylometry for Multilingual Manuscript. In IOSR Journal of Engineering, 2(9), pages 1–6. https://doi.org/10.9790/3021-02910106 A Appendices Features specifications: 1. Language-constant features: Where we had six features that are considered to be constant in the language, which are known and cannot be considered in terms of personal opinion. These features are: Part of Speech, Function word, Negation, and Punctuation. Filling these words was based on previous knowledge of Arabic grammar. In addition, the features (Abbreviation and Emoji) took either the value 0 or 1 value based on the existence of the feature in the word. The following provide more details and examples about each feature and explanation of why they were considered as constant features. 2. Features that carry a yes/no answer. There were 34 features that only took values of 0, 1, or NA; the values cannot be a number in between 0-1. These features are: (Human, Alive, Female, Animal, Body-part, Food-related, Time-related, Place-related, Eaten, Past, Work/study, Question, Nationality, Weather, Prayer, Compare, Place, Time, Number, Many, Media-content, Listing, Quoting, Calling, Sport, Art, Policy, Literature, Religion, Science, Travel, Economy, Law, and Technology). These features are either applicable on the word or not, so they would take either the value 1 or 0. 3. The other features. The remaining 60 features may carry range of different values (e.g., 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, or NA) based on the word’s relatedness to the features. These features are: (Formal, Word correctness, Long, Strong, Expensive, Dangerous, Ability, Shyness, Like, Peaceful, Loyalty, Excellence, Privacy, Necessity, Reality, Desire, Request, Rich, Big, Normal, Beautiful, Smart, Useful, Cause-death, Noisy, Cold, Heavy, Thankful, Youthful, Romantic, Agree, Happy, Angry, Welcome, Sarcastic ,Similar, Scary, Disgusting, Well-known, Crime, Childish, Optimism, Simple, Comfortable, Interesting, Healthy, Surprised, Wonder, Argue, Certainty, Emphasis, Honesty, Emotional , Laugh, Abusive, Racist, Shame, Wish, and Royal). [↗](https://doi.org/10.9790/3021-02910106)

*Updated 1 November 2026*