💻 computer science

Does online sustainability communication shape public discourse? Insights from six years of tenant-housing provider interactions

This study analyzes six years of Dutch public housing interactions on Facebook to demonstrate that organizational characteristics, rather than post design, primarily shape the quality and structure of online sustainability discourse, with larger associations eliciting more substantive tenant engagement.

Original authors: Shray Juneja, Suzan Verberne, Ioulia V. Ossokina

Published 2026-07-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Shray Juneja, Suzan Verberne, Ioulia V. Ossokina

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine social media as a giant, noisy town square where organizations shout out updates about fixing up old houses to save energy. For a long time, the people in charge of these town squares only counted how many people waved back or gave a "thumbs up" (likes). They thought, "More waves mean more success!" But this study suggests that counting waves is like judging a whole pizza just by how many slices are left; it tells you how much people reacted, but not what they were actually saying.

The researchers decided to peek behind the curtain. They looked at 792 posts and 3,197 comments from 92 Dutch housing groups between 2018 and 2023. Instead of just counting, they used a clever computer "detective" to sort every single comment into six specific personality types:

On-topic feedback: "Hey, I like this plan."
On-topic criticism: "I hate this plan."
On-topic praise: "Great job!"
Off-topic complaints: "This plan is bad, and also my cat is sick."
Content sharing: "Look at this!" (just passing it along).
Information seeking: "Wait, how does this work?"

The Big Surprise: The Message Isn't the Boss
You might think that if a housing group writes a super-long, exciting post with a question at the end, people will start having deep, meaningful conversations. The paper explicitly argues against this idea. The researchers found that the design of the post (how long it was, if it had a link, or if it asked a question) didn't really change the type of conversation happening. It's like trying to change the flavor of a soup by changing the spoon you stir it with; the spoon doesn't matter as much as the ingredients already in the pot.

The Real Boss: The Organization Itself
So, what actually shapes the conversation? The paper suggests that the organization's own characteristics are the real architects of the discourse.

Size Matters: Bigger housing groups (the "giants" of the neighborhood) tended to get more serious, question-filled, and critical comments. It seems bigger organizations invite more substantive engagement.
Rent Prices: Groups with lower rents (more affordable housing) got fewer critical or evaluative comments. The data suggests that tenants in these groups were less likely to launch into deep debates or complaints compared to others.
Happiness Levels: When tenants were already happy with their housing, they were less likely to write long, emotional, or questioning comments. Instead, they tended to just passively share the posts. It's as if a happy tenant thinks, "Everything is fine, I'll just share this with my friend," whereas a less happy tenant might write a whole essay about why things need to change.

The "Town Square" Effect
The study also found that the housing groups were actually pretty good at keeping the conversation on track. When they posted about sustainability, the tenants' replies were much more focused on that topic than if the comments had been randomly mixed up with other posts. This suggests that even in a chaotic digital world, the organization still holds the steering wheel of the conversation.

What's New and What's Still a Mystery
The researchers didn't just guess; they used a machine-learning pipeline to sort thousands of comments and ran statistical tests to see if these patterns were real. They found that over time, especially after 2020, people started asking more questions and offering more criticism.

However, the paper is careful to say this is an observational study. They watched what happened but didn't run an experiment where they changed the posts to see what would happen. So, while the results strongly suggest that organizational size and rent levels drive the conversation, they aren't a "proven" cause-and-effect law just yet. They also admit that looking at just one platform (Facebook) and one type of housing might mean the results don't apply perfectly to every situation in the world.

In short, if you want to understand how people talk about sustainability online, don't just look at the post's font size or length. Look at who is posting, how big they are, and how happy their tenants are. The organization's identity shapes the conversation far more than the words on the screen.

Technical Summary: Online Sustainability Communication and Public Discourse

Problem Statement
Authorities increasingly utilize social media to drive sustainability transitions, infrastructure investment, and service reform. However, the mechanisms by which citizens respond to these digital communications remain poorly understood. Existing research predominantly relies on aggregate engagement metrics (e.g., likes, shares, comment counts), which capture participation volume but obscure the qualitative nature of discourse. These metrics fail to distinguish between distinct forms of engagement—such as critical evaluation, information seeking, or passive redistribution—which vary significantly in affect, communicative function, and alignment with the original message. Furthermore, current typologies are often qualitative, theoretically derived, or limited to single dimensions (e.g., sentiment or topic modeling), failing to capture the multidimensionality of discourse where affective orientation, communicative intent, and semantic relatedness interact. There is a lack of scalable methods to simultaneously analyze these dimensions to understand how organizational communication shapes public discourse.

Methodology
The authors developed a data-driven, multidimensional framework to analyze discourse configurations in the context of Dutch public housing associations. The study utilized a dataset of 792 posts and 3,197 tenant comments from the Facebook pages of 92 housing associations between 2018 and 2023. The methodology proceeded in three stages:

Operationalization of Three Dimensions:
- Semantic Relatedness: Measured as the cosine similarity between post and comment embeddings using the paraphrase-multilingual-MiniLM-L12-v2 model. A permutation test confirmed that comments were significantly more semantically aligned with their source posts than with randomly paired content ( $p < 0.001$ ).
- Communicative Intent: Classified into five categories (Appreciation, Criticism, Inquiry, Statement, Forwarding) based on speech act theory. Due to the low-resource nature of fine-grained Dutch intent data, the authors employed a data augmentation strategy using Large Language Models (LLMs) to generate 13,000 synthetic training examples. An ensemble of three fine-tuned Dutch transformer models (GroNLP/bert-base-dutch-cased, DTAI-KULeuven/robbert-2023-dutch-large, and MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) achieved 85% accuracy on a held-out test set via majority voting.
- Sentiment: Classified into Positive, Neutral, and Negative. An ensemble of three pre-trained models (nlptown/bert-base-multilingual-uncased-sentiment, clips/republic, and citizenlab/twitter-xlm-roberta-base-sentiment-finetunned) was used, achieving 72% accuracy on a hand-labeled test set.
Discourse Typology Construction:
The three dimensions were combined into a unified feature space (five intent probabilities, one normalized sentiment score, and one semantic relatedness score). Unsupervised $k$ -means clustering was applied to identify recurring configurations. The optimal number of clusters was determined to be six ( $k=6$ ) based on the silhouette score.
Statistical Analysis of Drivers:
A multinomial logistic regression model was employed to estimate the effects of post-level communication design (length, lexical diversity, question presence, URL presence) and organizational characteristics (housing stock scale, rent affordability, tenant satisfaction) on the resulting discourse types. The analysis focused on the top 50 organizations by comment count (648 posts, 2,882 comments).

Key Results

Discourse Structure: The analysis identified six distinct discourse types: on-topic feedback, on-topic criticism, on-topic praise, off-topic complaints, content sharing, and information seeking. These types represent structured configurations of intent, sentiment, and topical alignment, ranging from substantive engagement to passive redistribution.
Semantic Alignment: Tenant comments were significantly more semantically aligned with their corresponding posts than with random pairings, indicating that organizational communication effectively structures the topics of public response.
Temporal Dynamics: From 2020 onwards, there was a marked increase in on-topic criticism and information seeking, while content sharing and on-topic praise remained stable. This suggests a shift toward more evaluative and inquiry-driven engagement rather than a uniform increase in activity.
Drivers of Discourse:
- Organizational Characteristics: These were the dominant predictors. Larger housing associations attracted more substantive responses (criticism and inquiry). Organizations with lower rents (higher affordability) received fewer evaluative comments (feedback, criticism, complaints). Higher tenant satisfaction was associated with a relative suppression of expressive and inquiry-driven discourse, with tenants defaulting to passive content sharing.
- Communication Design: Post-level features (length, questions, URLs) did not significantly explain variation in discourse types once organizational context was controlled for. The structural properties of the posts did not translate into systematic differences in discourse composition.

Significance and Contributions
The paper makes two primary contributions:

Methodological: It introduces a scalable, data-driven framework that jointly models semantic relatedness, sentiment, and communicative intent. This approach enables the identification of discourse configurations that are invisible to conventional engagement metrics, providing a more fine-grained account of citizen responses.
Empirical: It demonstrates that in the context of public housing sustainability communication, discourse patterns are shaped primarily by structural organizational conditions (scale, affordability, satisfaction) rather than the design of individual messages.

The authors conclude that while organizations retain agenda-setting capacity (as evidenced by semantic alignment), the nature of public discourse is deeply embedded in the organizational context. The study challenges the reliance on aggregate metrics as proxies for public response quality, arguing for multidimensional analysis to understand legitimacy and trust in digital governance. The framework is presented as transferable to other organizational and policy settings for analyzing online discourse dynamics.

Limitations
The study acknowledges that its observational nature precludes causal claims regarding communication features and discourse outcomes. Generalizability may be limited by the focus on a single platform (Facebook) and policy domain (Dutch public housing). Additionally, the binarization of organizational characteristics and the interpretative choices in clustering may obscure nuanced relationships.

Technical Summary: Online Sustainability Communication and Public Discourse

More like this