Elamur Research⌕
☰
home / papers / 2026 / From Detection to Explanation: Modeling Fine-Grained Emotional Social Influence Techniques with LLMs and Human Preferences

From Detection to Explanation: Modeling Fine-Grained Emotional Social Influence Techniques with LLMs and Human Preferences

Paper · 2026 · Maciej Markiewicz, Wiktoria Mieleszczenko-Kowszewicz, Beata Bajcar et al.

Summary

LLMs increasingly shape opinions, and emotional pressure such as guilt, fear or excitement can steer people without their noticing. Earlier work mostly classified persuasion at the text level. This paper asks whether LLMs can find emotional social influence techniques inside dialogues and explain them in ways people find useful.

The authors built EmoSocInflu, a set of 238 Polish dialogues with span-level labels for 11 emotional techniques. It is derived from an earlier corpus, and three PhD psychologists reviewed the spans as superannotators.

  • Study 1: LLMs were prompted zero-shot at temperature 0 in two steps. First they named techniques with rationales. Then, for each correctly detected technique, they listed the matching spans, which were compared with the human annotations.
  • Study 2: Human pairwise preferences over model explanations were turned into an ELO ranking. Explanations were also rated for comprehensibility, cognitive coherence, completeness and soundness.

Current models are weak at fine-grained detection. GPT-4o locates spans best but classifies poorly. Claude 3.5 Sonnet classifies better and produces the most preferred explanations. The public release supports research on detecting manipulation and raising awareness of it.

Key points

  • Dataset: EmoSocInflu has 238 Polish dialogues. It was cut from 247 after super-annotation, which found 9 texts with none of the selected techniques. It covers 11 techniques, including Emotional see-saw, Guilt, Anticipatory regret and Power of word 'love'. The data is available at github.com/social-influence/emo-soc-influ.
  • Annotation: 11 annotators labelled the spans at sentence level, with 2 per text. Three psychologist superannotators then corrected the spans and filled in missing ones.
  • Models tested: GPT-4o, o3, o4-mini, Claude 3.5 Sonnet, Mixtral 8x22B Instruct and Llama 3.1 70B Instruct. Spans were requested as quoted text rather than token tags or indices.
  • Explanation quality: Completeness and soundness drove overall human preference most strongly, with completeness singled out as the most decisive. Claude 3.5 Sonnet was consistently preferred.
  • Limitations: Only 27 text–technique pairs were detected by all four compared LLMs. Some techniques, such as Take advantage of bad mood, are underrepresented. The data is Polish only, and no models were fine-tuned.

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

🎓Wrocław University of Science and TechnologyWroclaw · Poland8 works · 945 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksPhysics and Astronomy · 44 worksStatistical and Nonlinear Physics · 40 works

🧭 Topics 1Research subjects and languages the work is about.Sentiment & emotion · 972 works

Cite

Cite as Maciej Markiewicz, Wiktoria Mieleszczenko-Kowszewicz, Beata Bajcar, Tomasz Adamczyk, Aleksander Szczęsny, Jolanta Babiak and Przemyslaw Kazienko. 2026. From Detection to Explanation: Modeling Fine-Grained Emotional Social Influence Techniques with LLMs and Human Preferences. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-srw.53

@inproceedings{markiewicz2026detection,
  title = {{From Detection to Explanation: Modeling Fine-Grained Emotional Social Influence Techniques with LLMs and Human Preferences}},
  author = {Maciej Markiewicz, Wiktoria Mieleszczenko-Kowszewicz, Beata Bajcar, Tomasz Adamczyk, Aleksander Szczęsny, Jolanta Babiak and Przemyslaw Kazienko},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-srw.53},
  url = {https://doi.org/10.18653/v1/2026.eacl-srw.53}
}

References

31 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anthropic. 2024. Claude 3.5 Sonnet. https://docs.anthropic.com/en/docs/ about-claude/models/all-models. Proprietary License. ↗
  2. Hui Bai, Jan G Voelkel, Shane Muldowney, Johannes C Eichstaedt, and Robb Willer. 2025. LLM generated messages can persuade humans on policy issues. Nature Communications, 16(1):6037. \#\# Ethical considerations Bartosz Broda, Bartłomiej Nitoń, Włodzimierz Gruszczyński, and Maciej Ogrodniczuk. 2014.… find
  3. This work has clear research benefits but also carries dual-use risks, as the methods could potentially Matteo Bruno, Renaud Lambiotte, and Fabio Saracco. 2022. Brexit and bots: characterizing the behaviour of automated accounts on Twitter during the uk election. EPJ Data Science, 11(1):17. find
  4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  5. Arpad E. Elo. 1978. The Rating of Chessplayers, Past and Present. Arco Publishing, New York. find
  6. Maram Hasanain, Fatema Ahmad, and Firoj Alam. 2024a. Can GPT-4 identify propaganda? annotation and detection of propaganda spans in news articles. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 2724–… find
  7. Maram Hasanain, Fatema Ahmad, and Firoj Alam. 2024b. Large language models for propaganda span annotation. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 14522– 14532, Miami, Florida, USA. Association for Computational Linguistics. find
  8. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  9. Nazanin Jafari, James Allan, and Sheikh Muhammad Sarwar. 2024. Target span detection for implicit harmful content. In ICTIR 2024 - Proceedings of the 2024 ACM SIGIR International Conference on the Theory of Information Retrieval, pages 117–122. Association for Computing Machinery, Inc. find
  10. Zdeněk Kasner, Vilém Zouhar, Patrícia Schmidtová, Ivan Kartáč, Kristýna Onderková, Ondřej Plátek, Dimitra Gkatzia, Saad Mahamood, Ondřej Dušek, and Simone Balloccu. 2025. Large language models as span annotators. find
  11. Todd Kulesza, Simone Stumpf, Margaret Burnett, Sherry Yang, Irwin Kwan, and Weng-Keen Wong. 2013. Too much, too little, or just right? ways explanations impact end users’ mental models. In 2013 IEEE Symposium on visual languages and human centric computing, pages 3–10. IEEE. find
  12. Q Vera Liao and S Shyam Sundar. 2022. Designing for responsible trust in ai systems: A communication perspective. In Proceedings of the 2022 ACM conference on fairness, accountability, and transparency, pages 1257–1268. find
  13. Helena Löfström, Karl Hammar, and Ulf Johansson. 2022. A meta survey of quality evaluation criteria in explanation methods. In International Conference on Advanced Information Systems Engineering, pages 55–63. Springer. find
  14. Aniek F Markus, Jan A Kors, and Peter R Rijnbeek. 2021. The role of explainability in creating trustworthy artificial intelligence for health care: a comprehensive survey of the terminology, design choices, and evaluation strategies. Journal of biomedical informatics, 113:103655. find
  15. MetaAI. 2024. Meta Llama 3.1 70B Instruct. https://huggingface.co/meta-llama/Llama-3. 1-70B-Instruct. Llama 3.1 Community License. ↗
  16. Microsoft Threat Analysis Center. 2024. Same targets, new playbooks: East Asia threat actors employ unique methods. Technical report, Microsoft Security Insider. PDF available from Microsoft Threat Analysis Center. find
  17. Wiktoria Mieleszczenko-Kowszewicz, Beata Bajcar, Aleksander Szcz˛esny, Maciej Markiewicz, Jolanta Babiak, Berenika Dyczek, and Przemysław Kazienko. 2025. Unraveling SITT: Social influence technique taxonomy and detection with llms. In Proceedings of the SENTIRE Workshop at the IEEE International Con… find
  18. Tim Miller. 2019. Explanation in artificial intelligence: Insights from the social sciences. Artificial intelligence, 267:1–38.
  19. Abhika Mishra, Akari Asai, Vidhisha Balachandran, Yizhong Wang, Graham Neubig, Yulia Tsvetkov, and Hannaneh Hajishirzi. 2024. Fine-grained hallucination detection and editing for language models.
  20. Przemysław Kazienko and Erik Cambria. 2024. Toward responsible recommender systems. IEEE Intelligent Systems, 39(3):5–12. find
  21. Danush Khanna, Pratinav Seth, Sidhaarth Sredharan Murali, Aditya Kumar Guru, Siddharth Shukla, Tanuj Tyagi, Sandeep Chaurasia, and Kripabandhu Ghosh. 2025. SELF-PERCEPT: Introspection improves large language models’ detection of multi-person mental manipulation in conversations. In Proceedings of th… find
  22. MistralAI. 2024. Mixtral-8x22B-Instruct-v0.1: A sparse Mixture of Experts Language Model. https://huggingface.co/mistralai/ Mixtral-8x22B-Instruct-v0.1. Apache 2.0 License. ↗
  23. OpenAI. 2025. Openai api documentation: o3 and o4-mini. Accessed: 2025-12-12. find
  24. Nicolas Scharowski, Sebastian AC Perrig, Melanie Svab, Klaus Opwis, and Florian Brühlmann. 2023. Exploring the effects of human-centered ai explanations on trust and reliance. Frontiers in Computer Science, 5:1151150. find
  25. Johannes Schneider. 2024. Explainable generative AI (GenXAI): a survey, conceptualization, and research agenda. Artificial Intelligence Review, 57(11):289. find
  26. Aleksander Szcz˛esny, Maciej Markiewicz, Łukasz Radliński, and Przemysław Kazienko. 2025. Leveraging positional bias of LLM in-context learning with Class-Few-Shot and Maj-Min alternating ordering. In Computational Science – ICCS 2025: 25th International Conference, Singapore, Singapore, July 7–9,… find
  27. Joanna Szwoch, Mateusz Staszkow, Rafal Rzepka, and Kenji Araki. 2024. Limitations of large language models in propaganda detection task. Applied Sciences, 14(10). find
  28. Giulia Vilone and Luca Longo. 2021. Notions of explainability and evaluation approaches for explainable artificial intelligence. Information Fusion, 76:89–106. find
  29. Raúl Vázquez, Timothee Mickus, Elaine Zosa, Teemu Vahtola, Jörg Tiedemann, Aman Sinha, Vincent Segonne, Fernando Sánchez-Vega, Alessandro Raganato, Jindřich Libovický, Jussi Karlgren, Shaoxiong Ji, Jindřich Helcl, Liane Guillou, Ona de Gibert, Jaione Bengoetxea, Joseph Attieh, and Marianna Apidian… find
  30. David Wan, Koustuv Sinha, Srini Iyer, Asli Celikyilmaz, Mohit Bansal, and Ramakanth Pasunuru. 2024. ACUEval: Fine-grained hallucination evaluation and correction for abstractive summarization. In Findings of the Association for Computational Linguistics: ACL 2024, pages 10036–10056, Bangkok, Thailan… find
  31. Jianlong Zhou, Amir H Gandomi, Fang Chen, and Andreas Holzinger. 2021. Evaluating the quality of machine learning explanations: A survey on methods and metrics. Electronics, 10(5):593. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/markiewicz-2026-detection-explanation-modeling-fine-grained/