Sentiment spreads, but topics do not, in COVID-19 discussions within the Belgian Reddit community

TL;DR

This study tackles how COVID‑19 mitigation topics and sentiments spread in the Belgian Reddit community. It combines topic labeling and sentiment analysis on 655,642 posts from 2020–2022 and introduces the Smooth Internal Expressed Bounded Confidence (SIEBC) model to explain sentiment dynamics with internal and expressed states updated via a bounded‑confidence kernel. Key findings show that topic volumes follow external events rather than social contagion within Reddit, while sentiment demonstrates contagion and homophily within threads, with measured values such as $h = 0.228$, $0.198$, and $0.133$ and a Wasserstein distance range $W_1 o$ between $0.493$ and $0.607$ from observed to predicted sentiments. The work offers a mechanism to estimate hidden internal sentiment and suggests topic‑specific incorporation of sentiment signals into epidemic‑social models, providing practical guidance for more nuanced disease forecasting and policy evaluation.

Abstract

This study investigates how topics and sentiments on COVID-19 mitigation measures -- specifically lockdowns, mask mandates, and vaccinations -- spread through the Belgian Reddit community. We explore 655,642 posts created between 1 January 2020 and 30 June 2022. In line with previous studies for other countries and platforms, we find that the volume of posts on these topics can be tied to important external events, but not within-Reddit interactions. Sentiment, however, is influenced by the sentiment of previous posts, resulting in homophily and polarisation. We define a homophily measure and find values of 0.228, 0.198, and 0.133 for lockdowns, masks and vaccination, respectively. Additionally, we introduce a novel bounded confidence model that estimates internal sentiment of users from their expressed sentiment. The Wasserstein metric between the predicted and the observed sentiments takes values between 0.493 (vaccination) and 0.607 (lockdown). These results yield insight into the way the Belgian Reddit community experienced the pandemic, and which aspects influenced the topics discussed and their associated sentiment.

Figures (9)

• Figure 1: Structure of a Reddit thread. The parent and ancestors of the focal comment are shown. The initiating posts for the mask and vaccination discussions are denoted by † the others are participating (see Section \ref{['sec:met-activity']}).
• Figure 2: The distribution of posts made per user on a double logarithmic scale for the topic (A) lockdowns, (B) masks, and (C) vaccination.
• Figure 3: (A) Schematic representation of the Smooth Internal Expressed Bounded Confidence (SIEBC, Equation \ref{['eq:model']}). (B) The change in sentiment after applying $B_{\alpha, \epsilon}$ (Equation \ref{['eq:bc']}) as a function of the sentiment difference $s_2 - s_1$.
• Figure 4: The number of posts per day on (A) lockdowns, (C) masks, (D) vaccination, (B) the number of hospitalisations, and (E) the number vaccination doses in Belgium. All quantities are shown as their two-week rolling mean. Piecewise linear trends are given by dashed lines. Red vertical lines mark significantly negative days. Important events are denoted by vertical dashed lines or gray regions Table \ref{['tab:temporal']}.
• Figure 5: Proportion of users $\rho(i)$ that has at least one initiated post in the first $i$ discussions for the three topics, (A) lockdowns, (B) masks, and (C) vaccination. The full line represents the observed proportion, the dashed line the expected proportion under the null model.