Elamur Research⌕
☰
home / papers / 2022 / Multilevel sentiment analysis in arabic

Multilevel sentiment analysis in arabic

Paper · 2022 · Ahmed Nassar and Ebru Sezer

Summary

The paper tackles polarity classification (positive vs. negative) for Arabic text at two granularities: individual sentiment terms and whole documents. Arabic sentiment lexicons are scarce, and one word may carry several sense-specific (posterior) polarity scores. The authors therefore first look for a way to reduce those scores to a single prior polarity per lemma. They compare several aggregation formulas, some new to Arabic, and select the best one. The chosen term scores then feed feature vectors for both term-level and document-level classification. They also add hand-written rules for negation and intensification. Three standard classifiers are compared: SVM, ANN and Decision Tree. ANN wins at both levels, with an average F-score of 0.92 for terms and 0.94 (positive) / 0.93 (negative) for documents. The authors stress that their pipeline is interpretable and repeatable. It runs from term scores up to document polarity, which suits a language where purely lexicon-based methods are limited.

Key points

  • Two-class sentiment analysis (positive/negative) is done at term level and document level for Arabic.
  • Several formulas for merging posterior polarities into one prior polarity are tested. The one named M Max_Sub is recommended. The approach follows Guerini et al.'s work on SentiWordNet.
  • The classifiers compared are SVM (LIBSVM in Matlab, default parameters, RBF kernel), ANN and D-Tree. ANN is best at both levels.
  • Results: term-level average F-score is 0.92. Document-level F-score is 0.94 for positive and 0.93 for negative test classes.
  • Rules for negation and intensification are combined with the ML approach. Features are built both with and without these rules applied at the term level.

Interesting ideas

  • I1. When a lemma has several sense-specific polarity values, collapse them into one prior score with an explicit formula, and pick that formula by testing candidates on real data rather than by intuition; this gives one usable value even when the sense is unknown.
  • I2. Build document-level features out of the same term-level scores, so a document's verdict can be traced back to the terms that drove it instead of to opaque features.
  • I3. Layer small hand-written rules for negation and intensifiers on top of a learned classifier, so known linguistic operators flip or scale polarity deterministically.
  • I4. Compare the same feature set with and without the rule layer across several classifiers, which isolates what the rules contribute apart from the choice of learner.

For Mova

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

  • I1 (world): Compiled lemma-level evaluative values from WordNet senses could be aggregated this way, giving a documented fallback polarity when word-sense choice is uncertain.
  • I2 (frames): Stance or evaluation of a whole text could be composed from per-frame evaluative scores, keeping every step explainable as Mova requires.
  • I3 (frames): In Mova such rules could be scoped by the UD tree (e.g., negation or degree modifiers attached to the evaluative head) rather than by word windows, which would help irony and implicature handling.
  • I4 (learn): The same ablation design could be used when evaluating self-induced repair rules, giving a clean with/without comparison alongside negative controls.

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

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 608 worksSemitic languages · 1,484 works

Cite

Cite as Ahmed Nassar and Ebru Sezer. 2022. Multilevel sentiment analysis in arabic. arXiv:2205.12328.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2205.12328 · arXiv 2205.12328

@misc{nassar2022multilevel,
  title = {{Multilevel sentiment analysis in arabic}},
  author = {Ahmed Nassar and Ebru Sezer},
  year = {2022},
  doi = {10.48550/arxiv.2205.12328},
  eprint = {2205.12328},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2205.12328}
}

References

10 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Wikipedia, Sentiment analysis. https://en.wikipedia.org/wiki/Sentiment_analysis (accessed March, 2017). ↗
  2. Internet World Stats, Internet world users by language. http://www.internetworldstats.com/stats7.htm (accessed February 2017). ↗
  3. Medhat, Walaa, Ahmed Hassan, and Hoda Korashy, 2014. Sentiment analysis algorithms and applications: A survey. Ain Shams Engineering Journal, 5(4), 1093-1113. ↗
  4. Badaro, Gilbert, et al., 2014. A large scale Arabic sentiment lexicon for Arabic opinion mining. ANLP 2014, 165. find
  5. Mahyoub, Fawaz HH, Muazzam A. Siddiqui, and Mohamed Y. Dahab, 2014. Building an Arabic sentiment lexicon using semi-supervised learning. J King Saud Univ Inf Sci 2014;26:417–24. ↗
  6. Mohammad, Saif, Mohammad Salameh, and Svetlana Kiritchenko, 2016. Sentiment Lexicons for Arabic Social Media. In LREC. find
  7. Boudad, N., Faizi, R., Thami, R. O. H., & Chiheb, R. (2017). Sentiment analysis in Arabic: A review of the literature. Ain Shams Engineering Journal. ↗
  8. Liu, Bing, 2012. Sentiment analysis and opinion mining. Synthesis lectures on human language technologies, 5(1), 1-167. ↗
  9. Turney, Peter D., (2002, July). Thumbs up or thumbs down?: semantic orientation applied to unsupervised classification of reviews. In Proceedings of the 40th annual meeting on association for computational linguistics (pp. 417-424). Association for Computational Linguistics.
  10. Farra, N., Challita, E., Assi, R. A., & Hajj, H. (2010, December). Sentence-level and document-level sentiment mining for arabic texts. In Data Mining Workshops (ICDMW), 2010 IEEE International Conference on (pp. 1114-1119). IEEE. Analysis and Opinion Mining. In LREC (Vol. 10, pp. 2200-2204). [13] Rushdi‐Saleh, M., Martín‐Valdivia, M. T., Ureña‐López, L. A., & Perea‐ Ortega, J. M. (2011). OCA: Opinion corpus for Arabic. Journal of the American Society for Information Science and Technology, 62(10), 2045-2054. [14] El-Halees, Alaa, 2011. Arabic opinion mining using combined classification approach, the International Arab Conference on Information Technology, pp. 10–13. [15] Shoukry, Amira, and Ahmed Rafea, (2012, May). Sentence-level Arabic sentiment analysis. In Collaboration Technologies and Systems (CTS), 2012 International Conference on (pp. 546-550). IEEE. [16] Moraes, Rodrigo, JoãO Francisco Valiati, and Wilson P. GaviãO Neto, 2013. Document-level sentiment classification: An empirical comparison between SVM and ANN. Expert Systems with Applications, 40(2), 621-633. [17] Joachims, Thorsten, 1996. A Probabilistic Analysis of the Rocchio Algorithm with TFIDF for Text Categorization (No. CMU-CS-96-118). Carnegie-mellon univ pittsburgh pa dept of computer science. [18] Li, Yung-Ming, and Tsung-Ying Li, 2013. Deriving market intelligence from microblogs. Decision Support Systems, 55(1), 206-217. [19] Chang, Chih-Chung, and Chih-Jen Lin, 2011. LIBSVM: a library for support vector machines. ACM transactions on intelligent systems and technology (TIST), 2(3), 27. [20] Atakulreka, Akarachai, and Daricha Sutivong, (2007, December). Avoiding local minima in feedforward neural networks by simultaneous learning. In Australasian Joint Conference on Artificial Intelligence (pp. 100-109). Springer, Berlin, Heidelberg. [21] Quinlan, J. Ross, 1986. Induction of decision trees. Machine learning, 1(1), 81-106. [22] Jia, Lifeng, Clement Yu, and Wei Zhang, 2008. UIC at TREC 2008 blog track. ILLINOIS UNIV AT CHICAGO DEPT OF COMPUTER SCIENCE. [23] Wikipedia, Arabic. https://en.wikipedia.org/wiki/Arabic (accessed May 2017). [24] Abdelali, Ahmed, Jim Cowie, and Hamdy S. Soliman, (2004, April). Arabic information retrieval perspectives. In Proceedings of the 11th Conference on Natural Language Processing, Journes d’Etude sur la Parole-Traitement Automatique des Langues Naturelles (JEP-TALN) (pp. 391-400). [25] Benajiba, Yassine, and Paolo Rosso, (2007, June). Towards a measure for Arabic corpora quality. In Proc. Int. Colloquium on Arabic Language Processing, CITALA-2007, Rabat, Morroco. [26] Yahya, Adnan, and Ali Salhi, 2014. Quality Assessment of General and Categorized Arabic Text Corpora. Proceedings of 5th International Conference on Arabic Language Processing-CITALA-2014, Woujda, Morocco. [27] Maamouri, M., Graff, D., Bouziri, B., Krouna, S., Bies, A., & Kulick, S. (2010). Standard Arabic morphological analyzer (SAMA) version 3.1. Linguistic Data Consortium, Catalog No.: LDC2010L01. [28] Miller, G. A., Beckwith, R., Fellbaum, C., Gross, D., & Miller, K. J. (1990). Introduction to WordNet: An on-line lexical database. International journal of lexicography, 3(4), 235-244. [29] Black, W., Elkateb, S., Rodriguez, H., Alkhalifa, M., Vossen, P., Pease, A., & Fellbaum, C. (2006, January). Introducing the Arabic wordnet project. In Proceedings of the third international WordNet conference (pp. 295-300). [30] Kohavi, Ron, (1995, August). A study of cross-validation and bootstrap for accuracy estimation and model selection. In Ijcai (Vol. 14, No. 2, pp. 1137-1145). [31] Al-Azani, Sadam, and El-Sayed M. El-Alfy, 2017. Using Word Embedding and Ensemble Learning for Highly Imbalanced Data Sentiment Analysis in Short Arabic Text. Procedia Computer Science, 109, 359-366. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nassar-2022-multilevel-sentiment-analysis-arabic/