---
title: "Improving Sentiment Analysis in Arabic Using Word Representation"
description: "Sentiment classification for Arabic is difficult: the language has rich morphology, inconsistent spelling… 26 references, cited by 3 papers here."
authors: ["Abdulaziz M. Alayba", "Vasile Palade", "Matthew England", "Rahat Iqbal"]
year: 2018
venue: "arXiv:1803.00124 · Proc. 2nd International Workshop on Arabic and Derived Script Analysis and Recognition (ASAR '18), pp. 13-18. IEEE, 2018"
source: "https://arxiv.org/abs/1803.00124"
doi: "10.1109/asar.2018.8480191"
arxiv: "1803.00124"
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/alayba-2018-improving-sentiment-analysis-arabic-using/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/alayba-2018-improving-sentiment-analysis-arabic-using/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2018](https://md.research.elamur.ai/free/en/papers/2018/page/5/) / Improving Sentiment Analysis in Arabic Using Word Representation

# Improving Sentiment Analysis in Arabic Using Word Representation

**Authors:** [Abdulaziz M. Alayba](https://md.research.elamur.ai/free/en/authors/abdulaziz-m-alayba/), [Vasile Palade](https://md.research.elamur.ai/free/en/authors/vasile-palade/), [Matthew England](https://md.research.elamur.ai/free/en/authors/matthew-england/) and [Rahat Iqbal](https://md.research.elamur.ai/free/en/authors/rahat-iqbal/) · **Year:** 2018 · **Venue:** arXiv:1803.00124 · Proc. 2nd International Workshop on Arabic and Derived Script Analysis and Recognition (ASAR '18), pp. 13-18. IEEE, 2018 · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://doi.org/10.1109/asar.2018.8480191) · [PDF](https://arxiv.org/pdf/1803.00124) · [DOI 10.1109/asar.2018.8480191](https://doi.org/10.1109/asar.2018.8480191) · [arXiv 1803.00124](https://arxiv.org/abs/1803.00124) · [OpenAlex](https://openalex.org/W2789618396)

**Citations:** 56 citations (OpenAlex, 2026-10-14) · [3](https://md.research.elamur.ai/free/en/papers/alayba-2018-improving-sentiment-analysis-arabic-using/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Sentiment classification for Arabic is difficult: the language has rich morphology, inconsistent spelling (the Hamza letter is a frequent source of errors) and many dialects, and tweets are very short. The authors trained Word2Vec models with Gensim on the Abu El-Khair Corpus, about 1.5 billion words of news text from ten newspapers in eight Arab countries. They cleaned the corpus by removing non-Arabic tokens, digits and punctuation, and normalised several letter variants. They built CBOW and Skip-gram models at five sizes. They chose one by checking which words each model placed nearest to "good" and "bad", and settled on CBOW with 200 dimensions. That model then served two purposes. First, it grew an automatic sentiment lexicon that fed seven scikit-learn classifiers. Second, it supplied embeddings to a CNN that also takes lexicon features. Both setups were tested on the authors' Arabic Health Services tweet dataset. Accuracy rose to 0.92 on the full dataset, up from 0.90 in the authors' earlier work, and to 0.95 on a subset where all annotators agreed. The work shows that general-domain embeddings can help with small, domain-specific sentiment data.

## Key points
- **Corpus and preprocessing:** Abu El-Khair Corpus, XML_UTF-8 version.
  - After filtering it is 16.55 GB, and the word count dropped by only 0.31%.
  - Normalisation maps the alef variants (آ إ أ) to ا and maps ة to ه.
- **Embedding models:** CBOW and SG were each built at sizes 10, 50, 100, 200 and 300.
  - Selection was qualitative, based on the nearest neighbours of the probe words.
  - A model was rejected if an antonym or a sentiment-neutral word such as "natural/authentic" appeared among those neighbours.
  - CBOW-200 was the one selected.
- **Data:** the AHS dataset has 2026 tweets (628 positive, 1398 negative).
  - The Sub-dataset keeps only tweets on which all three annotators agreed: 1732 tweets, about 85% of the total.
  - Both are public on Bitbucket.
- **Classical classifiers:** MNB, BNB, NSVC, LSVC, LR, SGD and RDG.
  - Features were TF, TF-IDF, POS, a manual lexicon (Lex) and Auto-Lex.
  - POS tags came from Stanford CoreNLP and were limited to verbs and adjectives.
  - Results are ten-fold cross-validated, and LSVC and Ridge performed best.
- **CNN:** an adapted Naïve Concatenation model combining CNN and lexicon features.
  - It uses filter sizes 3/4/5, an 80/20 split and 100 epochs.
  - Two lexicons were tested, separately and together: the SemEval-2016 Arabic Twitter Lexicon (1366 entries, graded from −1 to +1) and a hand-built Arabic Health Twitter Lexicon (716 entries, ±1).
  - Using both lexicons gave the best results: 0.92 on the main set and 0.95 on the Sub-dataset.

## Interesting ideas
- The lexicon is expanded in two hops with a shrinking fan-out: ten neighbours of each seed word, then only five neighbours of each of those, followed by deduplication. The smaller second hop limits drift toward antonyms, a practical guard against embedding neighbours that mix opposite polarities.
- An embedding model is chosen by inspecting the neighbours of probe words. A model is rejected if an antonym or a sentiment-neutral word shows up near a polar seed, which is a cheap qualitative sanity test before any downstream training.
- The same experiments are reported on the full dataset and on a unanimous-agreement subset. The gap between the two shows how much of the error comes from label noise rather than from the model.
- Spelling variants that people commonly confuse (the Hamza and alef forms) are folded into one form before training, so that misspellings do not split a word's statistics.
- Part-of-speech features are restricted to verbs and adjectives, treating them as the main carriers of opinion instead of using every token.

## For Mova

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

- **I1** (world): when growing evaluative or value vocabularies from WordNet or embedding neighbours, keep the second hop small and log each word's origin, so every lexicon entry stays explainable.
- **I2** (learn): add "antonym-in-neighbours" and "neutral-word-near-seed" checks as negative controls whenever a similarity resource is admitted into the pipeline.
- **I3** (learn): score silver data separately on items where all witness parsers agree and on contested items, to tell parser weakness apart from annotation ambiguity.
- **I4** (tokens): a similar explicit, logged normalisation layer for English variants (apostrophe forms, British/American spellings) before lemmatisation would keep the lemma counts unified.
- **I5** (frames): evaluative frames and irony detection could look first at adjectival predicates and evaluative verbs in the UD tree, rather than scanning all words.

*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

- [Coventry University](https://md.research.elamur.ai/free/en/universities/coventry-university/) — University, Coventry, United Kingdom · 4 works · 411 citations

📚 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 (2): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 972 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Abdulaziz M. Alayba, Vasile Palade, Matthew England and Rahat Iqbal. 2018. Improving Sentiment Analysis in Arabic Using Word Representation. arXiv:1803.00124 · Proc. 2nd International Workshop on Arabic and Derived Script Analysis and Recognition (ASAR '18), pp. 13-18. IEEE, 2018.

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

```bibtex
@misc{alayba2018improving,
  title = {{Improving Sentiment Analysis in Arabic Using Word Representation}},
  author = {Abdulaziz M. Alayba, Vasile Palade, Matthew England and Rahat Iqbal},
  year = {2018},
  doi = {10.1109/asar.2018.8480191},
  eprint = {1803.00124},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1109/asar.2018.8480191}
}
```

- [← previous Enhancing Universal Dependency Treebanks: A Case Study](https://md.research.elamur.ai/free/en/papers/nivre-2018-enhancing-universal-dependency-treebanks-case/)
- [next → Annotation of Tense and Aspect Semantics for Sentential AMR](https://md.research.elamur.ai/free/en/papers/donatelli-2018-annotation-tense-aspect-semantics-sentential/)
- [↑ back to papers of 2018](https://md.research.elamur.ai/free/en/papers/2018/page/5/)

## References

26 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. A. Alayba, V. Palade, M. England, and R. Iqbal, “[Arabic Language Sentiment Analysis on Health Services](https://md.research.elamur.ai/free/en/papers/alayba-2017-arabic-language-sentiment-analysis-health/),” in 1st International Workshop on Arabic Script Analysis and Recognition (ASAR), 2017, pp. 114–118.
2. N. Y. Habash, “Introduction to Arabic Natural Language Processing,” Synth. Lect. Hum. Lang. Technol., vol. 3, no. 1, pp. 1–187, Jan. 2010. [↗](https://doi.org/10.1007/978-3-031-02139-8)
3. K. Shaalan, “Rule-based Approach in Arabic Natural Language Processing,” Int. J. Inf. Commun. Technol., vol. 3, no. 3, pp. 11–19, 2010. [find](https://scholar.google.com/scholar?q=K%2E+Shaalan%2C+%E2%80%9CRule%2Dbased+Approach+in+Arabic+Natural+Language+Processing%2C%E2%80%9D+Int%2E+J%2E+Inf%2E+Commun%2E+Technol%2E%2C+vol%2E+3%2C+no%2E+3%2C+pp%2E+11%E2%80%9319%2C+2010%2E)
4. M. Althobaiti, U. Kruschwitz, and M. Poesio, “[AraNLP: a Java-based Library for the Processing of Arabic Text](https://md.research.elamur.ai/free/en/papers/althobaiti-2014-aranlp-java-based-library-processing/).,” Proc. Ninth Int. Conf. Lang. Resour. Eval., pp. 4134–4138, 2014.
5. I. Abu El-khair, “[1.5 billion words Arabic Corpus](https://md.research.elamur.ai/free/en/papers/el-khair-2016-1-5-billion-words-arabic/),” arXiv:1611.04033[cs], Dec. 2016. [↗](https://arxiv.org/abs/1611.04033)
6. B. Pang, L. Lee, and S. Vaithyanathan, “[Thumbs up? Sentiment Classification using Machine Learning Techniques](https://md.research.elamur.ai/free/en/papers/pang-2002-thumbs-up-sentiment-classification-machine/),” in Proceedings of the ACL-02 conference on Empirical methods in natural language processing - EMNLP ’02, 2002, vol. 10, pp. 79–86. [↗](https://doi.org/10.48550/arxiv.cs/0205070)
7. T. Mullen and N. Collier, “Sentiment analysis using support vector machines with diverse information sources,” in Proceedings of EMNLP-04, 9th Conference on Empirical Methods in Natural Language Processing, 2004, pp. 412– 418. [find](https://scholar.google.com/scholar?q=T%2E+Mullen+and+N%2E+Collier%2C+%E2%80%9CSentiment+analysis+using+support+vector+machines+with+diverse+information+sources%2C%E2%80%9D+in+Proceedings+of+EMNLP%2D04%2C+9th+Conference+on+Empirical+Methods+in+Natural+Language+Proce)
8. M. Speriosu, N. Sudan, S. Upadhyay, and J. Baldridge, “Twitter Polarity Classification with Label Propagation over Lexical Links and the Follower Graph,” Proc. Conf. Empir. Methods Nat. Lang. Process., pp. 53–56, 2011. [find](https://scholar.google.com/scholar?q=M%2E+Speriosu%2C+N%2E+Sudan%2C+S%2E+Upadhyay%2C+and+J%2E+Baldridge%2C+%E2%80%9CTwitter+Polarity+Classification+with+Label+Propagation+over+Lexical+Links+and+the+Follower+Graph%2C%E2%80%9D+Proc%2E+Conf%2E+Empir%2E+Methods+Nat%2E+Lang%2E+Process%2E)
9. A. Farghaly and K. Shaalan, “Arabic natural language processing : Challenges and solutions,” ACM Trans. Asian Lang. Inf. Process., vol. 8, no. 4, pp. 1–22, 2009. [↗](https://doi.org/10.1145/1644879.1644881)
10. A. Shoukry and A. Rafea, “Sentence-level Arabic sentiment analysis,” in 2012 International Conference on Collaboration Technologies and Systems (CTS), 2012, pp. 546–550. [find](https://scholar.google.com/scholar?q=A%2E+Shoukry+and+A%2E+Rafea%2C+%E2%80%9CSentence%2Dlevel+Arabic+sentiment+analysis%2C%E2%80%9D+in+2012+International+Conference+on+Collaboration+Technologies+and+Systems+%28CTS%29%2C+2012%2C+pp%2E+546%E2%80%93550%2E)
11. M. Abdul-Mageed, M. Diab, and M. Korayem, “[Subjectivity and Sentiment Analysis of Modern Standard Arabic](https://md.research.elamur.ai/free/en/papers/mageed-2011-subjectivity-sentiment-analysis-modern-standard/),” Proc. 49th Annu. Meet. Assoc. Comput. Linguist. Hum. Lang. Technol., no. 3, pp. 587– 591, 2011.
12. A. Mourad and K. Darwish, “[Subjectivity and Sentiment Analysis of Modern Standard Arabic and Arabic Microblogs](https://md.research.elamur.ai/free/en/papers/mourad-2013-subjectivity-sentiment-analysis-modern-standard/),” Proc. 4th Work. Comput. Approaches to Subj. Sentim. Soc. Media Anal., no. 3, pp. 55– 64, 2013.
13. A. A. Al Sallab, R. Baly, and H. Hajj, “[Deep Learning Models for Sentiment Analysis in Arabic](https://md.research.elamur.ai/free/en/papers/sallab-2015-deep-learning-models-sentiment-analysis/),” pp. 9–17, 2015. [↗](https://doi.org/10.13140/rg.2.1.4537.4167)
14. N. Abdulla, “Automatic Lexicon Construction for Arabic Sentiment Analysis,” 2014. [find](https://scholar.google.com/scholar?q=N%2E+Abdulla%2C+%E2%80%9CAutomatic+Lexicon+Construction+for+Arabic+Sentiment+Analysis%2C%E2%80%9D+2014%2E)
15. T. Mikolov, K. Chen, G. Corrado, and J. Dean, “[Efficient Estimation of Word Representations in Vector Space](https://md.research.elamur.ai/free/en/papers/mikolov-2013-efficient-estimation-word-representations-vector/),” pp. 4–5, 2013. [↗](https://doi.org/10.48550/arxiv.1301.3781)
16. J. Pennington, R. Socher, and C. Manning, “[Glove: Global Vectors for Word Representation](https://md.research.elamur.ai/free/en/papers/pennington-2014-glove-global-vectors-word-representation/),” Proc. 2014 Conf. Empir. Methods Nat. Lang. Process., pp. 1532–1543, 2014.
17. I. Abu El-khair, “Ibrahim Abu El-Khair,” 2017. [Online]. Available: [http://www.abuelkhair.net/index.php/en/arabic/abu-el-khair-corpus](http://www.abuelkhair.net/index.php/en/arabic/abu-el-khair-corpus). [Accessed: 10-Apr-2017]. [↗](http://www.abuelkhair.net/index.php/en/arabic/abu-el-khair-corpus)
18. T. Mikolov, K. Chen, G. Corrado, and J. Dean, “[Efficient Estimation of Word Representations in Vector Space](https://md.research.elamur.ai/free/en/papers/mikolov-2013-efficient-estimation-word-representations-vector/),” pp. 1–12, 2013. [↗](https://doi.org/10.48550/arxiv.1301.3781)
19. R. Řehůřek, “models.word2vec – Deep learning with word2vec,” 2017. [Online]. Available: [https://radimrehurek.com/gensim/models/word2vec.html](https://radimrehurek.com/gensim/models/word2vec.html). [Accessed: 16-Feb-2017]. [↗](https://radimrehurek.com/gensim/models/word2vec.html)
20. H. Manning, C., Raghavan, P. and Schütze, Introduction to information retrieval, 1st Ed. New York: Cambridge University Press, 2008. [find](https://scholar.google.com/scholar?q=H%2E+Manning%2C+C%2E%2C+Raghavan%2C+P%2E+and+Sch%C3%BCtze%2C+Introduction+to+information+retrieval%2C+1st+Ed%2E+New+York%3A+Cambridge+University+Press%2C+2008%2E)
21. C. Manning, M. Surdeanu, J. Bauer, J. Finkel, S. Bethard, and D. McClosky, “[The Stanford CoreNLP Natural Language Processing Toolkit](https://md.research.elamur.ai/free/en/papers/manning-2014-stanford-corenlp-natural-language-processing/),” Proc. 52nd Annu. Meet. Assoc. Comput. Linguist. Syst. Demonstr., pp. 55–60, 2014.
22. scikit-learn developers, “scikit-learn,” 2010. . [find](https://scholar.google.com/scholar?q=scikit%2Dlearn+developers%2C+%E2%80%9Cscikit%2Dlearn%2C%E2%80%9D+2010%2E+%2E)
23. Y. Kim, “[Convolutional Neural Networks for Sentence Classification](https://md.research.elamur.ai/free/en/papers/kim-2014-convolutional-neural-networks-sentence-classification/),” Proc. 2014 Conf. Empir. Methods Nat. Lang. Process. (EMNLP 2014), pp. 1746–1751, 2014.
24. X. Wang, W. Jiang, and Z. Luo, “Combination of Convolutional and Recurrent Neural Network for Sentiment Analysis of Short Texts,” pp. 2428–2437, 2016. [find](https://scholar.google.com/scholar?q=X%2E+Wang%2C+W%2E+Jiang%2C+and+Z%2E+Luo%2C+%E2%80%9CCombination+of+Convolutional+and+Recurrent+Neural+Network+for+Sentiment+Analysis+of+Short+Texts%2C%E2%80%9D+pp%2E+2428%E2%80%932437%2C+2016%2E)
25. B. Shin, T. Lee, and J. D. Choi, “Lexicon Integrated CNN Models with Attention for Sentiment Analysis,” in Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, 2017, pp. 149–158. [find](https://scholar.google.com/scholar?q=B%2E+Shin%2C+T%2E+Lee%2C+and+J%2E+D%2E+Choi%2C+%E2%80%9CLexicon+Integrated+CNN+Models+with+Attention+for+Sentiment+Analysis%2C%E2%80%9D+in+Proceedings+of+the+8th+Workshop+on+Computational+Approaches+to+Subjectivity%2C+Sentiment+and+So)
26. S. Kiritchenko, S. M. Mohammad, and M. Salameh, “[SemEval-2016 task 7: Determining sentiment intensity of English and Arabic phrases](https://md.research.elamur.ai/free/en/papers/kiritchenko-2016-semeval-2016-task-7-determining/),” Proc. Int. Work. Semant. Eval., pp. 42–51, 2016.

*Updated 1 November 2026*