Dashboard – MUSMA PRIN2022

What you will find here

topic trends over time: how themes emerge, peak, and fade
sentiment & toxicity dynamics: how emotions and tone shift
polarization indicators: when and where conversations diverge

The analysis is powered by MUSMA models, designed to process millions of posts and reveal patterns that are invisible at smaller scales.

All information is based on the following publications [Storani25][Loru25]. All code is openly available at the project’s GitHub repository.

Table of Contents

Quick data statistics

Topic clustering

Sentiment

Hashtags

Hashtags and reliability

Hashtag dynamics

Web pages and websites

Engaging with sources

Polarization

Quick data statistics

21M

total posts

12M

from Twitter/X

8M

from Facebook

650k

from Instagram

38k

from YouTube

_
Posted between 01/01/2018 and 31/12/2022
Talking about: climate change

(Posts are analyzed in aggregate to capture global trends rather than individual behavior)

Topic clustering

This section maps the semantic structure of climate-related discussions by grouping posts into coherent topic clusters.

Using representation learning and clustering techniques, each post is embedded in a semantic space and grouped based on similarity of content, revealing major themes and sub-discourses within the conversation.

How to read the plot

Each point represents a post
Distance reflects semantic similarity (closer = more similar content)
Colors indicate automatically identified topic clusters
The 2D layout is a projection of a higher-dimensional semantic space

Clusters do not overlap conceptually, but may appear close in the projection due to dimensionality reduction.

What this view reveals

Core vs peripheral topics
Dense central regions correspond to high-volume, general discussions, while smaller, isolated clusters capture more specific or niche themes.
Topical diversity
The presence of many medium-sized clusters highlights the fragmentation of climate discourse into multiple concurrent narratives.
Bridging topics
Clusters positioned between others often act as semantic connectors, linking activism, policy, science, and news-oriented discussions.

⚠️ Interpretation note

Cluster labels and boundaries are data-driven and approximate. This visualization is intended to support exploration and comparison, not strict topic taxonomy.

Sentiment over time

How does the tone of the conversation change across platforms?

This heatmap shows the evolution of public sentiment over time for each social media platform. It highlights periods of heightened negativity or positivity, as well as platform-specific dynamics that may reflect differences in audience, moderation, or communication style.

How to read it

Color: average sentiment level (🔵 blue for positive, 🟡 yellow for neutral, 🔴 red for negative)
X-axis: time
Y-axis: social media platform

What stands out

Platform-specific tone differences
Instagram conversations are consistently more 🔵 positive than other platforms, while X (Twitter) shows a more 🔴 negative and volatile tone over time.
Temporal fluctuations align across platforms
Several sentiment shifts occur simultaneously across platforms, suggesting reactions to shared external events (e.g. policy announcements, climate reports, global crises).
Data availability matters
The absence of Twitter data starting from 2022 reflects collection or access limitations, not a change in sentiment. This highlights how platform policies can shape what can be analyzed.

⚠️ Note: Sentiment values reflect aggregate trends, not individual opinions. Differences across platforms may arise from audience composition, moderation practices, and posting behavior, in addition to the underlying public discourse.

The discourse in hashtags

This section highlights the most frequent hashtags used in climate-related conversations across social media platforms.

The treemaps below emphasize both dominant narratives and platform-specific framing.

How to read it

Box size reflects hashtag frequency
Color intensity reinforces relative prominence
Each treemap is platform-specific
(Instagram, YouTube, Twitter, Facebook)

Twitter/X

Facebook

Instagram

YouTube

Global

Key observations

Different modes of engagement
Facebook shows a more institutional and informational framing, whereas Twitter exhibits higher fragmentation and political signaling.
A shared core vocabulary
Hashtags like #climatechange, #climate, and #climateaction dominate across all platforms, indicating a common baseline discourse.
Platform-specific emphases
Instagram foregrounds activism and lifestyle themes (e.g. #sustainability, #earthday), while YouTube and Twitter surface more event-driven and news-oriented hashtags (e.g. #COP26, #news).

Hashtags and reliability

By comparing the hashtags used in posts linking to reliable and unreliable sources, we can observe how different narratives emerge within the climate change debate.

For each platform, we identified the most frequently used hashtags and compared their relative frequency in posts sharing reliable versus unreliable content. The figures below show the difference in usage for the most common hashtags appearing in posts from 🔴 unreliable and 🔵 reliable sources.

Key observations

The analysis highlights clear differences in how climate-related discussions are framed:

Reliable sources tend to use hashtags connected to scientific discussions and international policy events, such as
#cop26, #cop27, and #actionclimate.
Unreliable sources more frequently employ hashtags associated with controversial or sceptical narratives, including terms such as #climatehoax.

Platform-specific dynamics

While most platforms show similar patterns, Instagram displays a slightly different structure, where some hashtags associated with unreliable content overlap with lifestyle or community-driven discussions (e.g., vegan or environmental communities). Meanwhile, hashtags used in reliable posts sometimes include geographically specific references, indicating links to regional climate initiatives or local discussions.

A polarized vocabulary

Overall, the differences in hashtag usage reveal a diverging vocabulary between the two groups. Posts linking to unreliable sources tend to employ hashtags that question or challenge the scientific consensus on climate change, while posts linking to reliable sources more often emphasize scientific evidence, policy discussions, and collective climate action.

Hashtag dynamics over time

While general hashtags remain consistently dominant, event-driven hashtags display sharp, short-lived spikes, reflecting reactions to external events such as climate summits or major news cycles.