Elamur Research⌕
☰
home / papers / 2007 / TAT: An Author Profiling Tool with Application to Arabic Emails

TAT: An Author Profiling Tool with Application to Arabic Emails

Paper · 2007 · Dominique Estival, Tanja Gaustad, Son Bao Pham et al.

Summary

The paper presents the Text Attribution Tool (TAT), a system built to infer characteristics of an email's writer rather than the writer's identity. It focuses on Arabic, a language with almost no prior profiling work. The authors gathered emails from 1030 native speakers of Egyptian Arabic. Each participant also filled in a questionnaire that supplied the target labels: three demographic traits (age, gender, education) and four psychometric traits from a customised EPQR-S. Arabic emails come in both Arabic and Latin ("franco-arabic") script, so the authors built dedicated preprocessing:

  • language and script detection,
  • separation of authored text from quoted or signature material,
  • tokenisation that handles multi-character Latin codes for a single Arabic glyph,
  • normalisation to Buckwalter transliteration.

Classifiers such as SMO and Bagging were then trained on grouped features. Six of seven traits beat the majority baseline with statistical significance. Gender reached 81.15% against a 72.16% baseline, and the psychometric gains were similar to those obtained for English. The results suggest that profiling an author is a realistic fallback when identifying an Arabic author is hard.

Key points

  • The Arabic corpus contains 8,028 emails (7,267 in Arabic script, 761 in Latin script) and about 2.15M words. The comparable English corpus has 1,033 authors and 9,836 emails.
  • Arabic writers had to contribute at least 2000 words, twice the English minimum, to compensate for the data sparseness caused by Arabic morphology.
  • Accuracy against the majority baseline for each trait:
    • gender: 81.15 vs 72.16
    • age: 72.10 vs 70.09
    • education: 93.66 vs 93.62 (no real gain, because the data is heavily skewed)
    • lie: 52.30 vs 40.41
    • neuroticism: 54.93 vs 43.42
    • psychoticism: 56.98 vs 49.39
    • extraversion: 54.35 vs 48.27
  • Features were organised into four groups: arabicNamedEntities, arabicChar, arabicMorphological and arabicLexical. Dropping arabicLexical lowered gender accuracy by 3.71 points, the largest loss of any group.
  • Bagging improved with InfoGain feature selection, whereas SMO did not. Significance was checked with a χ² test at p=0.01.

Interesting ideas

  • I1. Each processing module writes its output as stand-off annotations into a shared Annotation Repository, and a module's behaviour depends only on its input, so modules can be recombined freely.
  • I2. Text in several scripts or spelling conventions is first mapped to one unambiguous canonical encoding, and some multi-character sequences are read as a single unit.
  • I3. Leave-one-group-out ablation measures how much each feature family contributes to each trait, which makes the classifier's behaviour easier to interpret.
  • I4. Every trait is reported next to its own majority baseline, and the authors state plainly that a 93% score on skewed data says almost nothing.
  • I5. Gender agreement on a predicate with a first-person subject is treated as direct evidence about the speaker's gender.

For Mova

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

  • I1 (tokens): Tokenizer, tagger and lemma outputs could be stored as separate stand-off layers over the raw text, so later stages can always trace an analysis back to exact character spans.
  • I2 (tokens): Informal English spellings could be normalised into a canonical layer while the original form is kept, so the tagger sees clean tokens without the surface text being lost.
  • I3 (learn): Mova could disable one class of repair rules or one witness parser at a time and record the accuracy drop, giving an explainable account of what each component adds.
  • I4 (answer): Each school test or QA set should be scored against a "most frequent answer" control, so that high accuracy on lopsided tests is not mistaken for understanding.
  • I5 (frames): Coreference could link "I" in quoted speech to a speaker and attach any grammatical or lexical cues about that speaker as attributes of the referent.

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

🏢GoogleMountain View · United States430 works · 424,363 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksPsychology · 1,552 worksSocial Psychology · 306 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Dominique Estival, Tanja Gaustad, Son Bao Pham, Will Radford and Ben Hutchinson. 2007. TAT: An Author Profiling Tool with Application to Arabic Emails. Proceedings of the Australasian Language Technology Workshop 2007.

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

@inproceedings{estival2007tat,
  title = {{TAT: An Author Profiling Tool with Application to Arabic Emails}},
  author = {Dominique Estival, Tanja Gaustad, Son Bao Pham, Will Radford and Ben Hutchinson},
  booktitle = {Proceedings of the Australasian Language Technology Workshop 2007},
  year = {2007},
  url = {https://aclanthology.org/U07-1006/}
}

References

35 references, 5 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. 2005a. Applying authorship analysis to Arabic web content. In Paul B. Kantor et al., editor, Intelligence and Security Informatics, Proceedings of the IEEE International Conference on Intelligence and Security Informatics (ISI 2005), pages 183–197. Springer. find
  2. Ahmed Abbasi and Hsinchun Chen. 2005b. Applying authorship analysis to extremist-group web forum messages. IEEE Intelligent Systems, 20(5):67–75. ↗
  3. Ahmed Abdel-Khalek and Sybil Eysenenck. 1983. A cross-cultural study of personality: Egypt and England. Research in Behaviour and Personality, 3:215– 226. find
  4. David Aha, Dennis Kibler, and Mark Albert. 1991. Instance-based learning algorithms. Machine Learning, 6(1):37–66. ↗
  5. Shlomo Argamon, Sushant Dhawle, Mosche Koppel, and James W. Pennebaker. 2005. Lexical predictors of personality type. In Proceedings of the 2005 Joint Annual Meeting of the Interface and the Classification Society of North America, St. Louis. find
  6. Leo Breiman. 1996. Bagging predictors. Learning, 24(2):123–140.
  7. Machine Leo Breiman. 2001. Random forests. Machine Learning, 45(1):5–32. find
  8. Tom Buchanan, John A. Johnson, and Lewis R. Goldberg. 2005. Implementing a Five-Factor personality inventory for use on the internet. European Journal of Psychological Assessment, 21:115–127. ↗
  9. Tim Buckwalter. 2000. Arabic transliteration. http://www.qamus.org/transliteration.htm. ↗
  10. Tim Buckwalter. 2002. Arabic morphological analyzer. Linguistic Data Consortium (LDC2002L49). find
  11. Tim Buckwalter. 2004. Issues in Arabic orthography and morphology analysis. In Proceedings of the COLING 2004 Workshop on computational approaches to Arabic script-based languages, Geneva.
  12. William Cohen. 1995. Fast effective rule induction. In Twelfth International Conference on Machine Learning, pages 115–123. ↗
  13. Malcolm Corney, Olivier de Vel, Alison Anderson, and George Mohay. 2002. Gender-preferential text mining of e-mail discourse. In Proceedings of the 18th Annual Computer Security Applications Conference (AC-SAC 2002), pages 282–292, Las Vegas. find
  14. Kareem Darwish. 2002. Building a shallow Arabic morphological analyzer in one day. In Proceedings of ACL 2002 Workshop on computational approaches to Semitic languages, Philadelphia. find
  15. Anne de Roeck and Waleed Al-Fares. 2000. A morphologically sensitive clustering algorithm for identifying Arabic roots. In Proceedings of ACL 2000, Toulouse.
  16. Olivier de Vel, Alison Anderson, Malcolm Corney, and George Mohay. 2001. Mining email content for author identification forensics. SIGMOD Record, 30(4):55–64. find
  17. Olivier de Vel, Alison Anderson, Malcolm Corney, and George Mohay. 2002. E-mail authorship attribution for computer forensics. In Daniel Barbara and Sushil Jajodia, editors, Data Mining for Security Applications. Kluwer Academic Publishers. ↗
  18. Olivier de Vel. 2000. Mining e-mail authorship. In Proceedings of the Workshop on Text Mining, ACM International Conference on Knowledge Discovery and Data Mining (KDD’2000), Boston. find
  19. Dominique Estival, Tanja Gaustad, Son Bao Pham, Will Radford, and Ben Hutchinson. 2007. Author profiling for English emails. In Proceedings of the 10th Conference of the Pacific Association for Computational Linguistics (PACLING 2007), pages ??–??, Melbourne. find
  20. Hans Eysenck and Sybil Eysenck. 1975. Manual of the Eysenck Personality Questionnaire. Hooder and Stoughton Educational, London. find
  21. Leslie Francis, Christopher Lewis, and Hans-Georg Ziebertz. 2006. The short-form revised Eysenenck personality questionnaire (EPQR-S): A German edition. Social Behaviour and Personality, 34(2):197– 204. find
  22. Yoav Freund and Robert Schapire. 1996. Experiments with a new boosting algorithm. In Thirteenth International Conference on Machine Learning, pages 148– 156, San Francisco. find
  23. Richard Friedmann. 1997. Harper, San Francisco. Who wrote the Bible? find
  24. Natalie Glance, Matthew Hurst, Kamal Nigam, Mathew Siegler, Robert Stockton, and Takashi Tomokiyo. 2005. Deriving marketing intelligence from online discussion. In Proceedings of the eleventh ACM SIGKDD international conference on Knowledge discovery in data mining, pages 419–428, Chicago. find
  25. Abduelbaset Goweder and Anne De Roeck. 2001. Assessment of a significant Arabic corpus. In Proceedings of the ACL/EACL Workshop on Arabic Language Processing: Status and Prospects, Toulouse. find
  26. Sathiya Keerthi, Shirish Shevade, C. Bhattacharyya, and K. Murthy. 2001. Improvements to Platt’s SMO algorithm for SVM classifier design. Neural Computation, 13(3):637–649. ↗
  27. Moshe Koppel, Jonathan Schler, and Kfir Zigdon. 2005. Automatically determining an anonymous author’s native language. In Paul B. Kantor et al., editor, Intelligence and Security Informatics, Proceedings of the IEEE International Conference on Intelligence and Security Informatics (ISI 2005), pages 209–217. Springer. find
  28. Gerrard Ledger and Thomas Merriam. 1994. Shakespeare, Fletcher, and The Two Noble Kinsmen. Literary and Linguistic Computing, 9(3):235–248. ↗
  29. F. Mosteller and D.L. Wallace. 1964. Inference and Disputed Authorship: The Federalist. Series in behavioral science: Quantitative methods edition. Addison-Wesley. find
  30. Warren T. Norman. 1963. Toward an adequate taxonomy of personality attributes: replicated factor structure in peer nomination personality rating. Journal of Abnormal and Social Psychology, 66:574–583. find
  31. Jon Oberlander and Scott Nowson. 2006. Whose thumb is it anyway? Classifying author personality from weblog text. In Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions, pages 627– 634, Sydney. find
  32. Ross Quinlan. 1993. C4.5: Programs for Machine Learning. Morgan Kaufmann, San Mateo.
  33. Ian H. Witten and Eibe Frank. 2005. Data Mining: Practical machine learning tools and techniques. Morgan Kaufmann, San Francisco, 2nd edition. find
  34. Rong Zheng, Yi Qin, Zan Huang, and Hsinchun Chen. 2003. Authorship analysis in cybercrime investigation. In Hsinchun Chen et al., editor, Proceedings of the first NSF/NIJ Intelligence and Security Informatics Symposium, pages 59–73. Springer. find
  35. Rong Zheng, Jiexun Li, Hsinchun Chen, and Zan Huang. 2006. A framework for authorship identification of online messages: Writing-style features and classification techniques. JASIST, 57(3):378–393. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/estival-2007-tat-author-profiling-tool-application/