Aligning Large Language Models with Human Opinions through Persona Selection and Value--Belief--Norm Reasoning
This work addresses the problem of inaccurate opinion prediction in AI for applications like social science or personalized systems, though it appears incremental as it builds on existing persona-based methods.
The paper tackles the challenge of aligning large language models with human opinions by addressing sensitivity to irrelevant personas and lack of strategic reasoning, proposing Chain-of-Opinion (COO) which improves opinion prediction by up to 4% and fine-tuned models by up to 23%.
Reasoning and predicting human opinions with large language models (LLMs) is essential yet challenging. Current methods employ role-playing with personae but face two major issues: LLMs are sensitive to even a single irrelevant persona, skewing predictions by up to 30%, and LLMs fail to reason strategically over personae. We propose Chain-of-Opinion (COO), a simple four-step solution modeling which and how to reason with personae, inspired by the Value--Belief--Norm (VBN) theory. COO differentiates between explicit personae (demographics and ideology) and implicit personae (historical opinions), involves: (1) filtering irrelevant attributes from explicit personae, (2) ranking implicit personae into a preferential list for selecting top-k, (3) applying novel VBN reasoning to extract user environmental and personal value, belief, and norm variables for accurate and reliable predictions, and (4) iterating VBN reasoning with progressively larger lists of implicit personae to handle potential persona insufficiency. COO efficiently achieves new state-of-the-art opinion prediction via prompting with only 5 inference calls, improving prior techniques by up to 4%. Notably, fine-tuning LMs with COO data results in significantly better opinion-aligned models, by up to 23%.