Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Latent Topic Synthesis: Leveraging LLMs for Electoral Ad Analysis

Alexander Brady, Tunazzina Islam

TL;DR

The paper tackles scalable analysis of political ads on social media by automatically constructing an interpretable topic taxonomy from unlabeled corpora using a density-based clustering backbone augmented by iterative LLM-driven labeling without seed sets. It combines embedding-based clustering (UMAP-HDBSCAN) with two-pass LLM topic synthesis and constrained decoding to generate coherent topics and moral framings, followed by cluster representative labeling and optional SetFit-based supervised classification. A case study on Meta ads from October 2024 demonstrates a dynamic taxonomy of political issues, associated moral foundations, and demographic targeting patterns, revealing that voting and immigration drive spending while abortion and election integrity achieve outsized reach. The approach outperforms BERTopic baselines in annotation quality, enables detailed analysis of funders, moral framing, and microtargeting, and provides a scalable, interpretable tool for researchers, policymakers, and the public to understand evolving digital political communication and polarization.

Abstract

Social media platforms play a pivotal role in shaping political discourse, but analyzing their vast and rapidly evolving content remains a major challenge. We introduce an end-to-end framework for automatically generating an interpretable topic taxonomy from an unlabeled corpus. By combining unsupervised clustering with prompt-based labeling, our method leverages large language models (LLMs) to iteratively construct a taxonomy without requiring seed sets or domain expertise. We apply this framework to a large corpus of Meta (previously known as Facebook) political ads from the month ahead of the 2024 U.S. Presidential election. Our approach uncovers latent discourse structures, synthesizes semantically rich topic labels, and annotates topics with moral framing dimensions. We show quantitative and qualitative analyses to demonstrate the effectiveness of our framework. Our findings reveal that voting and immigration ads dominate overall spending and impressions, while abortion and election-integrity achieve disproportionate reach. Funding patterns are equally polarized: economic appeals are driven mainly by conservative PACs, abortion messaging splits between pro- and anti-rights coalitions, and crime-and-justice campaigns are fragmented across local committees. The framing of these appeals also diverges--abortion ads emphasize liberty/oppression rhetoric, while economic messaging blends care/harm, fairness/cheating, and liberty/oppression narratives. Topic salience further reveals strong correlations between moral foundations and issues. Demographic targeting also emerges. This work supports scalable, interpretable analysis of political messaging on social media, enabling researchers, policymakers, and the public to better understand emerging narratives, polarization dynamics, and the moral underpinnings of digital political communication.

Latent Topic Synthesis: Leveraging LLMs for Electoral Ad Analysis

TL;DR

The paper tackles scalable analysis of political ads on social media by automatically constructing an interpretable topic taxonomy from unlabeled corpora using a density-based clustering backbone augmented by iterative LLM-driven labeling without seed sets. It combines embedding-based clustering (UMAP-HDBSCAN) with two-pass LLM topic synthesis and constrained decoding to generate coherent topics and moral framings, followed by cluster representative labeling and optional SetFit-based supervised classification. A case study on Meta ads from October 2024 demonstrates a dynamic taxonomy of political issues, associated moral foundations, and demographic targeting patterns, revealing that voting and immigration drive spending while abortion and election integrity achieve outsized reach. The approach outperforms BERTopic baselines in annotation quality, enables detailed analysis of funders, moral framing, and microtargeting, and provides a scalable, interpretable tool for researchers, policymakers, and the public to understand evolving digital political communication and polarization.

Abstract

Social media platforms play a pivotal role in shaping political discourse, but analyzing their vast and rapidly evolving content remains a major challenge. We introduce an end-to-end framework for automatically generating an interpretable topic taxonomy from an unlabeled corpus. By combining unsupervised clustering with prompt-based labeling, our method leverages large language models (LLMs) to iteratively construct a taxonomy without requiring seed sets or domain expertise. We apply this framework to a large corpus of Meta (previously known as Facebook) political ads from the month ahead of the 2024 U.S. Presidential election. Our approach uncovers latent discourse structures, synthesizes semantically rich topic labels, and annotates topics with moral framing dimensions. We show quantitative and qualitative analyses to demonstrate the effectiveness of our framework. Our findings reveal that voting and immigration ads dominate overall spending and impressions, while abortion and election-integrity achieve disproportionate reach. Funding patterns are equally polarized: economic appeals are driven mainly by conservative PACs, abortion messaging splits between pro- and anti-rights coalitions, and crime-and-justice campaigns are fragmented across local committees. The framing of these appeals also diverges--abortion ads emphasize liberty/oppression rhetoric, while economic messaging blends care/harm, fairness/cheating, and liberty/oppression narratives. Topic salience further reveals strong correlations between moral foundations and issues. Demographic targeting also emerges. This work supports scalable, interpretable analysis of political messaging on social media, enabling researchers, policymakers, and the public to better understand emerging narratives, polarization dynamics, and the moral underpinnings of digital political communication.
Paper Structure (24 sections, 6 figures, 4 tables)

This paper contains 24 sections, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Iterative Topic Synthesis
  5. Cluster Representative Labeling
  6. Supervised Classification
  7. Case Study
  8. Dataset
  9. Framework Application
  10. Moral Foundations
  11. Supervised Classification
  12. Demographic Targeting Analysis
  13. Results and Analysis
  14. Baseline Comparison
  15. Spending and Reach by Topic
  16. ...and 9 more sections

Figures (6)

  • Figure 1: Framework overview. All ads are initially embedded and clustered. (a) Each cluster is sequentially processed by the LLM to extend the topic taxonomy. (b) The generated topics are used to label the ads in each cluster.
  • Figure 2: Average spend and impressions reveal which topics dominate the ad landscape.
  • Figure 3: Top five funding entities by total spend across economy, abortion, and crime/justice topics. Economic ads are dominated by conservative groups, abortion spending reflects polarized investments from both anti-abortion and pro-abortion rights organizations, and crime/justice advertising is driven by a mix of PACs and local campaigns.
  • Figure 4: Topic-specific moral narratives: multiple frames shape economic ads, Liberty/Oppression dominates abortion related ads.
  • Figure 5: Correlations between moral foundations and topics. Color intensity indicates the strength of the correlation.
  • ...and 1 more figures