Bluesky user behavior prediction method wins SocialSim challenge

Researchers have developed a hybrid methodology to predict user actions on the social media platform Bluesky, addressing both common and rare behaviors. The approach combines historical response patterns, persona-specific LightGBM models for frequent actions, and a specialized neural network for rare action classification. This method achieved a macro F1-score of 0.64 for common actions and 0.56 for rare actions, demonstrating the need for tailored strategies based on action type. The work secured first place in the SocialSim challenge at the COLM 2025 workshop. AI

IMPACT This research offers a novel approach to understanding and predicting user actions on social media, potentially improving content recommendation systems and platform design.

RANK_REASON Academic paper detailing a new methodology for social media behavior prediction. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv cs.CL →

• Anastasia Shimorina
• arXiv
• Bluesky
• COLM 2025
• Hugging Face
• LightGBM
• SocialSim: Social-Media Based Personas

• paper
• product
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →