Synthesizing Public Opinions with LLMs: Role Creation, Impacts, and the Future to eDemorcacy

TL;DR

The paper tackles the problem of declining response rates and nonresponse bias in public opinion polling by introducing role creation based on knowledge injection, leveraging HEXACO-based personality profiles and demographic data to dynamically generate prompts via retrieval-augmented generation (RAG). Opinions are generated through conditioning, formalized as $p(x_n|x_1, \\ldots, x_{n-1})$, enabling nuanced, diverse simulated responses across demographic groups. Evaluated on 30 CES questions, the approach outperforms standard few-shot prompting, with larger models showing greater gains and adherence reaching up to ~84% in some configurations. The work discusses impacts, limitations, and future research directions toward cost-effective, scalable eDemocracy, including model-agnostic applicability and the need for ongoing data and compliance considerations.

Abstract

This paper investigates the use of Large Language Models (LLMs) to synthesize public opinion data, addressing challenges in traditional survey methods like declining response rates and non-response bias. We introduce a novel technique: role creation based on knowledge injection, a form of in-context learning that leverages RAG and specified personality profiles from the HEXACO model and demographic information, and uses that for dynamically generated prompts. This method allows LLMs to simulate diverse opinions more accurately than existing prompt engineering approaches. We compare our results with pre-trained models with standard few-shot prompts. Experiments using questions from the Cooperative Election Study (CES) demonstrate that our role-creation approach significantly improves the alignment of LLM-generated opinions with real-world human survey responses, increasing answer adherence. In addition, we discuss challenges, limitations and future research directions.

Figures (4)

• Figure 1: Taxonomy of LLM task customization approaches.
• Figure 2: Adapting LLMs to synthesizing public opinion tasks.
• Figure 3: Role generation from attributes.
• Figure 4: Opinion generation using roles.