Paid Voices vs. Public Feeds: Interpretable Cross-Platform Theme Modeling of Climate Discourse

Samantha Sudhoff, Pranav Perumal, Zhaoqing Wu, Tunazzina Islam · 2026

Read Paid Voices vs. Public Feeds: Interpretable Cross-Platform Theme Modeling of Climate Discourse by Samantha Sudhoff, Pranav Perumal, Zhaoqing Wu, Tunazzina Islam (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-13317.txt.

Introduction

Samantha Sudhoff, Pranav Perumal, Zhaoqing Wu, and Tunazzina Islam set paid climate ads next to free social posts and read both with one lens. Talk on climate shapes what the public grasps and what laws get made. But that talk runs on two kinds of stage with split pay rules.

Paid ad stages reward aimed, planned pleas. Free feed stages host raw talk by users at large. The team builds a theme model that can be read by eye. It maps which themes rule each stage and where they meet or part. For climate talk staff, the split shows whose voice a theme is, not just that it trends.

Catalog record

CREATORSamantha Sudhoff, Pranav Perumal, Zhaoqing Wu, Tunazzina Islam
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE73 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-13317

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Paid Voices vs. Public Feeds: Interpretable Cross-Platform Theme Modeling of Climate Discourse Samantha Sudhoff, Pranav Perumal, Zhaoqing Wu, Tunazzina Islam Abstract Climate discourse online plays a crucial role in shaping public understanding of climate change and influencing political and policy outcomes. However, climate communication unfolds across structurally distinct platforms with fundamentally different incentive structures: paid advertising ecosystems incentivize targeted, strategic persuasion, while public social media platforms host largely organic, user-driven discourse. Existing computational studies typically analyze these environments in isolation, limiting our ability to distinguish institutional messaging from public expression. In this work, we present a comparative analysis of climate discourse across paid advertisements on Meta (previously known as Facebook) and public posts on Bluesky from July 2024 to September 2025. We introduce an interpretable, end-to-end thematic discovery and assignment framework that clusters texts by semantic similarity and leverages large language models (LLMs) to generate concise, human-interpretable theme labels. We evaluate the quality of the induced themes against traditional topic modeling baselines using both human judgments and an LLM-based evaluator, and further validate their semantic coherence through downstream stance prediction and theme-guided retrieval tasks. Applying the resulting themes, we characterize systematic differences between paid climate messaging and public climate discourse and examine how thematic prevalence shifts around major political events. Our findings show that platform-level incentives are reflected in the thematic structure, stance alignment, and temporal responsiveness of climate narratives. While our empirical analysis focuses on climate communication, the proposed framework is designed to support comparative narrative analysis across heterogeneous communication environments. 1 Introduction Online platforms have become a primary arena for climate communication, shaping how climate change is discussed, debated, and acted upon by the public Pera and Aiello 2024 ; Islam et al. 2023b ; Bloomfield and Tillery 2019 ; Pearce et al. 2019 ; Abbar et al. 2016 . Climate narratives circulated through social media influence political attitudes, policy preferences, and perceptions of scientific credibility Liu et al. 2025 ; Vivion et al. 2024 ; Holder et al. 2023 ; Treen et al. 2020 ; Lejano and Nero 2020 ; Walter et al. 2018 ; Gupta et al. 2018 , making climate discourse a central object of study for computational social science (CSS) and natural language processing (NLP). Figure 1: Examples of climate discourse across paid and public platforms (paraphrased and anonymized). Importantly, climate communication does not occur in a uniform online environment. Instead, it unfolds across platforms with fundamentally different incentive structures. Paid advertising ecosystems, such as Meta’s advertising platform Meta Platforms, Inc. 2025 , incentivize strategic and targeted persuasion, where messages are carefully crafted to influence specific audiences and advance political or economic goals Islam et al. 2023b ; Islam et al. 2023a ; Islam and Goldwasser 2022 ; Capozzi et al. 2021 ; Ribeiro et al. 2019 . In contrast, public social media platforms such as Bluesky Bluesky Public Social Network 2025 host largely organic, user-driven discourse shaped by personal expression, social interaction, and grassroots mobilization. Recent work on emerging decentralized platforms such as Bluesky suggests that their design and user migration patterns produce distinct discourse dynamics compared to centralized platforms Quelle and Bovet 2024 . In this work, we treat paid advertising platforms as proxies for institutionally produced, resource-backed messaging, and public social media platforms as proxies for largely organic, user-driven discourse. These structural differences give rise to distinct communicative regimes that shape not only what is discussed, but how climate issues are framed. Fig. 1 illustrates how climate and energy issues are framed differently in paid advertisements and public social media posts. Paid platforms predominantly feature institutional actors promoting or legitimizing specific energy positions, while public platforms exhibit a wider range of individual perspectives, including both advocacy and denial. These contrasts motivate a systematic, cross-platform analysis of climate discourse that goes beyond within-platform studies. Fig. 1 provides illustrative examples of discourse framing and is intended solely as motivation; all empirical claims are derived from the analyses presented in later sections. Prior computational work has examined climate discourse within individual platforms, focusing on tasks such as topic modeling Dahal et al. 2019 , stance detection Upadhyaya et al. 2023 ; Luo et al. 2020 , and targeted messaging analysis Islam and Goldwasser 2025b ; Islam and Goldwasser 2025c ; Islam and Goldwasser 2025a ; Islam et al. 2023b . However, analyzing platforms in isolation obscures an essential question: to what extent observed narratives exhibit systematic patterns consistent with institutional messaging versus organic public discourse? Without a comparative lens, it is difficult to disentangle the effects of platform incentives from the content of climate communication itself. Figure 2: Overview of our framework. In this paper, we address this gap by conducting a comparative analysis of climate discourse across paid advertisements on Meta and public posts on Bluesky. We introduce an interpretable thematic discovery and assignment framework that enables direct comparison across platforms despite differences in vocabulary, style, and communicative norms. Our approach first discovers latent semantic clusters and then uses large language models (LLMs) Brown et al. 2020 to generate concise, human-interpretable theme labels, explicitly decoupling theme discovery from theme interpretation, allowing themes to remain grounded in the data while still being accessible for analysis and downstream use. This design avoids reliance on pre-defined topic inventories or seed sets, and enables robust comparison despite differences in surface language and framing. A high-level overview of our method is illustrated in Fig. 2 . We evaluate the quality of the resulting themes against standard topic modeling baselines, including Latent Dirichlet Allocation (LDA) Blei et al. 2003 and BERTopic Grootendorst 2022 , using both human judgments and an LLM-based evaluator. We then apply the resulting themes to characterize systematic differences between paid climate messaging and public climate discourse and to examine how thematic prevalence shifts around major political events. Our goal is not to attribute intent or causality to individual actors, but to characterize how platform-level incentives are reflected in the aggregate thematic structure of climate discourse. Additionally, to validate that the discovered themes capture meaningful semantic structure beyond surface-level clustering, we evaluate their utility on two downstream tasks: stance prediction and theme-guided retrieval . In the stance prediction task, we assess whether theme-based representations improve classification performance when used alone or in combination with text. In the retrieval task, we use themes as semantic queries to retrieve relevant texts across platforms, measuring how well themes capture semantically aligned content. These tasks serve as validation probes rather than end goals, providing evidence that the induced themes function as coherent and informative semantic abstractions. In summary, this paper makes three primary contributions 1 1 1 Our datasets and code are publicly available here: https://github.com/srs827/PaidVoices-vs-PublicFeeds . (1) We construct and release a multi-platform dataset enabling direct comparison between paid climate advertising and public social media discourse across Meta and Bluesky. (2) We propose an interpretable thematic modeling framework that explicitly decouples semantic discovery from theme interpretation, enabling robust cross-platform comparison despite differences in vocabulary and communicative norms. (3) We empirically demonstrate how platform-level incentives are reflected in thematic structure, stance alignment, and temporal responsiveness, and show that the resulting themes improve performance on downstream stance prediction and semantic retrieval tasks. 2 Related Work Climate and political campaigns in paid

Read the complete text