Researchers develop prompt tuning technique to reduce social attribution bias in large language models for behavioral analysis.

LLMs trained on human text may implicitly replicate attribution theory biases, attributing behavior to personal traits over situational factors

Study investigates incorporating user goals and message context into prompts to improve dispositional and situational causality reasoning

Chain-of-Thought reasoning paradigms risk producing biased responses when social attribution factors are ignored

New scalable method enriches instruction prompts with social context knowledge to mitigate these biases in online behavior analytics