Elamur Research⌕
☰
home / papers / 2026 / Predicting Social Media User Actions: A Hybrid Approach for Common and Rare Behavior Prediction on Bluesky

Predicting Social Media User Actions: A Hybrid Approach for Common and Rare Behavior Prediction on Bluesky

Paper · 2026 · Benjamin White and Anastasia Shimorina

Summary

The paper addresses predicting how a social media user will react to a conversation thread. Prior work has mostly studied frequent reactions such as likes and reposts, while rarer actions have received little attention. Working with BluePrint, a persona-based Bluesky dataset released for the SocialSim 2025 challenge, the authors (Orange Research) tackle a 12-action vocabulary. Each user belongs to one of 25 persona clusters, derived from embeddings of their posts. Their system combines four parts:

  • a lookup table of historical response patterns, used for high-confidence cases;
  • LightGBM classifiers trained per persona, with engineered temporal and semantic features, for frequent actions;
  • a dedicated neural model that merges text and timing representations to handle infrequent, imbalanced classes;
  • an LLM that writes the reply text.

The tabular models reach a mean macro F1 of 0.64 on common actions, and the rare-action model scores 0.56 macro F1. The work ranked first in the challenge. It suggests that different action types call for distinct modelling choices rather than a single LLM-only pipeline.

Key points

  • Data: BluePrint (Bück-Kaeffer et al., 2025), with 6,435,348 training and 213,556 development threads, 12 Bluesky actions and 25 persona clusters. Reported scores come from the development split, and the authors say private-test results were similar.
  • For common actions, persona-specific LightGBM with handcrafted temporal and semantic features beat transformer-based alternatives, reaching an average macro F1 of 0.64.
  • A hybrid text-plus-temporal neural architecture classifies 10 rare actions, reaching 0.56 macro F1.
  • Reply generation prompts GPT-4.1-mini with a liberal Bluesky-user persona and a 300-character limit.
  • Among all entries to the Social-Media Based Personas challenge (SocialSim), organized at COLM 2025's Social Simulation with LLMs workshop, the system was ranked top.

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

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksInformation Systems · 1,730 works

Cite

Cite as Benjamin White and Anastasia Shimorina. 2026. Predicting Social Media User Actions: A Hybrid Approach for Common and Rare Behavior Prediction on Bluesky. Proceedings of the 1st Workshop on Social Context (SoCon) and the 2nd Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/247khgfza8dd

@inproceedings{white2026predicting,
  title = {{Predicting Social Media User Actions: A Hybrid Approach for Common and Rare Behavior Prediction on Bluesky}},
  author = {Benjamin White and Anastasia Shimorina},
  booktitle = {Proceedings of the 1st Workshop on Social Context (SoCon) and the 2nd Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) @ LREC 2026},
  year = {2026},
  doi = {10.63317/247khgfza8dd},
  url = {https://doi.org/10.63317/247khgfza8dd}
}

References

9 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dimosthenis Antypas, Asahi Ushio, Francesco Barbieri, Leonardo Neves, Kiamehr Rezaee, Luis Espinosa-Anke, Jiaxin Pei, and Jose Camacho-Collados. 2023. SuperTweetEval: A Challenging, Unified and Heterogeneous Benchmark for Social Media NLP Research. In Findings of the Association for Computational Li… find
  2. Tadas Baltrušaitis, Chaitanya Ahuja, and Louis-Philippe Morency. 2018. Multimodal machine learning: A survey and taxonomy. IEEE transactions on pattern analysis and machine intelligence, 41(2):423–443. find
  3. Aurélien Bück-Kaeffer, Je Qin Chooi, Dan Zhao, Maximilian Puelma Touzel, Kellin Pelrine, Jean-François Godbout, Reihaneh Rabbany, and Zachary Yang. 2025. BluePrint: A Social Media User Dataset for LLM Persona Evaluation and Training. find
  4. Syeda Nadia Firdaus, Chen Ding, and Alireza Sadeghian. 2021. Retweet Prediction based on Topic, Emotion and Personality. Online Social Networks and Media, 25:100165. find
  5. Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Zhongyi Qiu, Hanjia Lyu, Wei Xiong, and Jiebo Luo. 2025. Can LLMs Simulate Social Media Engagement? A Study on Action-guided Response Generation. find
  6. Giulio Rossetti, Massimo Stella, Rémy Cazabet, Katherine Abramski, Erica Cau, Salvatore Citraro, Andrea Failla, Riccardo Improta, Virginia Morini, and Valentina Pansanella. 2024. Y Social: an LLM-powered Social Media Digital Twin. CoRR, abs/2408.00818. — [arXiv:2408.00818](https://arxiv.org/abs/2408.00818) find
  7. Konstantinos Sotiropoulos, John W Byers, Polyvios Pratikakis, and Charalampos E Tsourakakis. 2019. Twittermancer: predicting user interactions on twitter. In 2019 57th Annual Allerton Conference on Communication, Control, and Computing (Allerton), pages 973–980. IEEE. find
  8. Bongwon Suh, Lichan Hong, Peter Pirolli, and Ed H. Chi. 2010. Want to be Retweeted? Large Scale Analytics on Factors Impacting Retweet in Twitter Network. In Proceedings of the 2010 IEEE Second International Conference on Social Computing, SOCIALCOM ’10, page 177–184, USA. IEEE Computer Society. find
  9. Chenkai Sun, Jinning Li, Hou Pong Chan, ChengXiang Zhai, and Heng Ji. 2023. Measuring the effect of influential messages on varying personas. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 554–562, Toronto, find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/white-2026-predicting-social-media-user-actions/