Elamur Research⌕
☰
home / papers / 2022 / Emoji Sentiment Roles for Sentiment Analysis: A Case Study in Arabic Texts

Emoji Sentiment Roles for Sentiment Analysis: A Case Study in Arabic Texts

Paper · 2022 · Shatha Ali A. Hakami, Robert Hendley and Phillip Smith

Summary

The paper asks how emoji change the sentiment of the Arabic text they appear in, given that one emoji can push a message in different directions depending on context. The authors pooled 14 public Arabic Twitter datasets (144,196 tweets) into an Emoji-Text dataset. They then stripped all emoji to get a matching Plain-Text dataset. Both versions were labelled by four automatic annotators: Mazajak, CAMeL Tools, ASAD, and a lexicon-based Logit-scale scorer built on an augmented Ar-SeLn word lexicon plus the Arab-ESL emoji lexicon. Comparing each text's label with and without its emoji gives one of seven "sentiment states." Relating that state to the emoji's own polarity gives one of eleven "sentiment roles": emphasizer, indicator, mitigator, reverser or trigger of positive or negative sentiment, plus a neutral no-effect role. All eleven roles were observed in the data. A close study of "Face With Tears of Joy" showed that a nominally positive emoji can act in all three sentiment norms. This warns against treating emoji as fixed-polarity features.

Key points

  • The data comes from 14 Arabic Twitter datasets (144,196 tweets) and contains 1,034 unique emoji arranged in 24,364 distinct emoji patterns.
  • There are four annotators: Mazajak (CNN+LSTM, online API), the CAMeL Tools default sentiment model (fine-tuned AraBERT), the ASAD web API, and a lexicon-based Logit-scale scorer (Lowe et al., 2011).
  • The lexicon scorer combines Ar-SeLn, extended with dataset words to 178,620 entries, with Arab-ESL (1,034 emoji), and it yields intensity scores rather than bare labels.
  • The seven states are Keep, Reverse-to and Add for each polarity, plus a Neutral-State. These map onto eleven roles, and all eleven were found in the dataset.
  • "Face With Tears of Joy" is positive in Arab-ESL. Used alone, it appeared 421 times as a negativity mitigator in negative texts. Alongside negative emoji it took on negative roles linked to sarcasm, bullying, complaining and regret.

Interesting ideas

  • I1. Measure a feature's effect by building a minimal pair: run the same text through the annotators once with the feature and once with it deleted, then read the change in label as the feature's contribution.
  • I2. Define a role by crossing two things: how the text's label changed (kept, reversed, added, neutralised) and the feature's own polarity. This separates emphasizers, mitigators, reversers and triggers instead of just calling a cue positive or negative.
  • I3. Use several heterogeneous classifiers, and add a transparent lexicon scorer because the off-the-shelf tools give labels only, never intensities.
  • I4. A cue's role depends on how many times it occurs and on what co-occurs with it. Repeating the same positive emoji raised its weight enough to turn it from an emphasizer of negativity into a mitigator, because the stronger-intensity element dominates the pattern.
  • I5. Analyse one frequent, ambiguous item in depth and break its behaviour down by load, co-occurrence pattern and intensity. This exposes context-dependence that corpus-level averages hide.

For Mova

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

  • I1 (frames): Mova could delete one discourse marker, emoji or punctuation mark and re-run the irony/speech-act analysis, so that each pragmatic judgement can be explained by what that one token changed.
  • I2 (frames): the same grid could type valence shifters such as "barely" or "hardly" in the implicature layer, recording whether a modifier strengthens, softens or flips the evaluation in a frame.
  • I3 (learn): when silver labels are built from several witnesses, at least one witness should be a fully inspectable scorer that gives graded values, so disagreements can be traced to specific tokens.
  • I4 (frames): irony detection could treat repeated laughter or exclamation markers as cumulative evidence weighed against the literal polarity of the clause, rather than as a single binary flag.
  • I5 (learn): evaluation could include per-item deep dives on high-frequency ambiguous words, acting as targeted controls on whether the system's handling of a word changes with its context.

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

🎓University of BirminghamBirmingham · United Kingdom25 works · 62,564 citations🎓Jazan UniversityGazan · Saudi Arabia9 works · 98 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksHuman-Computer Interaction · 141 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 972 worksSemitic languages · 2,908 works

Cite

Cite as Shatha Ali A. Hakami, Robert Hendley and Phillip Smith. 2022. Emoji Sentiment Roles for Sentiment Analysis: A Case Study in Arabic Texts. 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. DOI 10.18653/v1/2022.wanlp-1.32

@inproceedings{hakami2022emoji,
  title = {{Emoji Sentiment Roles for Sentiment Analysis: A Case Study in Arabic Texts}},
  author = {Shatha Ali A. Hakami, Robert Hendley and Phillip Smith},
  booktitle = {Proceedings of the Seventh Arabic Natural Language Processing Workshop (WANLP)},
  year = {2022},
  doi = {10.18653/v1/2022.wanlp-1.32},
  url = {https://doi.org/10.18653/v1/2022.wanlp-1.32}
}

References

26 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shatha Ali A. Hakami, Robert Hendley, and Phillip Smith. 2022. A context-free Arabic emoji sentiment lexicon (CF-Arab-ESL). In Proceedinsg of the 5th Workshop on Open-Source Arabic Corpora and Processing Tools with Shared Tasks on Qur’an QA and Fine-Grained Hate Speech Detection, pages 51–59, Marseille, France. European Language Resources Association.
  2. Ibrahim Abu Farha and Walid Magdy. 2019. Mazajak: An online Arabic sentiment analyser. In Proceedings of the Fourth Arabic Natural Language Processing Workshop, pages 192–198, Florence, Italy. Association for Computational Linguistics.
  3. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. AraBERT: Transformer-based model for Arabic language understanding. In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection, pages 9–15, Marseille, France. European Language Resource Association. ↗
  4. Gilbert Badaro, Ramy Baly, Hazem Hajj, Nizar Habash, and Wassim El-Hajj. 2014. A large scale arabic sentiment lexicon for arabic opinion mining. In Proceedings of the EMNLP 2014 workshop on arabic natural language processing (ANLP), pages 165–173.
  5. Janet Beavin Bavelas and Nicole Chovil. 2000. Visible acts of meaning: An integrated message model of language in face-to-face dialogue. Journal of Language and social Psychology, 19(2):163–194. find
  6. John Carey. 1980. Paralanguage in computer mediated communication. In 18th Annual Meeting of the Association for Computational Linguistics, pages 67–69. find
  7. Daantje Derks, Arjan ER Bos, and Jasper Von Grumbkow. 2007. Emoticons and social interaction on the internet: the importance of social context. Computers in human behavior, 23(1):842–849. find
  8. Daantje Derks, Arjan ER Bos, and Jasper Von Grumbkow. 2008. Emoticons and online message interpretation. Social Science Computer Review, 26(3):379– 388. ↗
  9. Vyvyan Evans. 2017. The emoji code: The linguistics behind smiley faces and scaredy cats. Picador USA. find
  10. Tina Ganster, Sabrina C Eimler, and Nicole C Krämer. 2012. Same same but different!? the differential influence of smilies and emoticons on person perception. Cyberpsychology, Behavior, and Social Networking, 15(4):226–230. ↗
  11. Irving John Good. 1965. The estimation of probabilities: An essay on modern bayesian methods, vol. 30. find
  12. Shatha Ali A. Hakami, Robert Hendley, and Phillip Smith. 2020. Emoji as sentiment indicators: An investigative case study in arabic text. In The Sixth International Conference on Human and Social Analytics, pages 26–32. IARIA. find
  13. Sabit Hassan, Hamdy Mubarak, Ahmed Abdelali, and Kareem Darwish. 2021. Asad: Arabic social media analytics and understanding. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations, pages 113–118.
  14. Susan Herring and Ashley Dainas. 2017. “nice picture comment!” graphicons in facebook comment threads. In Proceedings of the 50th Hawaii International Conference on System Sciences. find
  15. Tianran Hu, Han Guo, Hao Sun, Thuy-vy Thi Nguyen, and Jiebo Luo. 2017. Spice up your chat: the intentions and sentiment effects of using emojis. In Eleventh international aaai conference on web and social media. find
  16. Sara Kiesler, Jane Siegel, and Timothy W McGuire. 1984. Social psychological aspects of computer-mediated communication. American psychologist, 39(10):1123. ↗
  17. Petra Kralj Novak, Jasmina Smailović, Borut Sluban, and Igor Mozetič. 2015. Sentiment of emojis. PloS one, 10(12):e0144296. ↗
  18. Shao-Kang Lo. 2008. The nonverbal communication functions of emoticons in computer-mediated communication. Cyberpsychology & behavior, 11(5):595–597. ↗
  19. Will Lowe, Kenneth Benoit, Slava Mikhaylov, and Michael Laver. 2011. Scaling policy preferences from coded political texts. Legislative studies quarterly, 36(1):123–155. ↗
  20. Tainyi Ted Luor, Ling-ling Wu, Hsi-Peng Lu, and Yu-Hui Tao. 2010. The effect of emoticons in simplex and complex task-oriented communication: An empirical study of instant messaging. Computers in Human Behavior, 26(5):889–895. ↗
  21. Mary L McHugh. 2012. Interrater reliability: the kappa statistic. Biochemia medica: Biochemia medica, 22(3):276–282. ↗
  22. Hannah Miller, Daniel Kluver, Jacob Thebault-Spieker, Loren Terveen, and Brent Hecht. 2017. Understanding emoji ambiguity in context: The role of text in emoji-related miscommunication. In Proceedings of the International AAAI Conference on Web and Social Media, volume 11. find
  23. Ossama Obeid, Nasser Zalmout, Salam Khalifa, Dima Taji, Mai Oudah, Bashar Alhafni, Go Inoue, Fadhl Eryani, Alexander Erdmann, and Nizar Habash. 2020. Camel tools: An open source python toolkit for arabic natural language processing. In Proceedings of the 12th language resources and evaluation conference, pages 7022–7032.
  24. Livia Polanyi and Annie Zaenen. 2006. Contextual valence shifters. In Computing attitude and affect in text: Theory and applications, pages 1–10. Springer. ↗
  25. Joseph B Walther and Kyle P D’addario. 2001. The impacts of emoticons on message interpretation in computer-mediated communication. Social science computer review, 19(3):324–347. ↗
  26. Philipp Wicke and Marianna Bolognesi. 2020. Emoji-based semantic representations for abstract and concrete concepts. Cognitive processing, 21(4):615– 635. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/hakami-2022-emoji-sentiment-roles-sentiment-analysis/