Can Large Language Models Accurately Predict Public Opinion? A Review - Archives ouvertes de la Science politique
Pré-Publication, Document De Travail Année : 2024

Can Large Language Models Accurately Predict Public Opinion? A Review

Arnault Pachot
  • Fonction : Auteur correspondant
  • PersonId : 1104041
  • IdHAL : arnault-pachot

Connectez-vous pour contacter l'auteur
Thierry Petit
  • Fonction : Auteur

Résumé

This article reviews the capacity of Large Language Models (LLMs) to accurately predict public opinion. Through a meta-analysis of recent research, we evaluate metrics such as accuracy, correlation, and bias to assess the implications for survey-based social science. Our findings indicate that LLMs demonstrate high accuracy in predicting demographic responses and aligning with survey data, suggesting their potential in reflecting public opinion trends. Despite this, challenges persist, particularly with regard to biases in race, gender, and ideology, which can distort model outputs. Several studies propose methods to measure and mitigate these biases, aiming to improve the representativeness of LLM predictions. By addressing these limitations and refining the ethical use of LLMs, these models could significantly enhance public opinion research, offering valuable insights for researchers, policymakers, and decision-makers in forecasting societal sentiments.
Fichier principal
Vignette du fichier
Can_Large_Language_Models_Accurately_Predict_Public_Opinion__A_Review__2024_09_10.pdf (556.84 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04688498 , version 1 (10-09-2024)

Identifiants

  • HAL Id : hal-04688498 , version 1

Citer

Arnault Pachot, Thierry Petit. Can Large Language Models Accurately Predict Public Opinion? A Review. 2024. ⟨hal-04688498⟩
15 Consultations
8 Téléchargements

Partager

More