Stable Emotional Co-occurrence Patterns Revealed by Network Analysis of Social Media

TL;DR

This work addresses how online emotion co-occurrence structures evolve in population-scale social media during crises versus normal times. It introduces a network-based, data-driven framework that maps posts to six POMS emotion dimensions via an expanded dictionary, derives signficant co-occurring concept links, and builds a higher-level emotion network whose link strengths are analyzed over time. The study reveals a remarkably stable rank order of emotion links across time and crises, with context-driven shifts in link strengths (notably Tension during earthquakes and pre-vaccination, Fatigue after vaccination), suggesting an intrinsic architecture of collective emotion. The approach offers a scalable tool for monitoring public sentiment and informing mental health interventions using large-scale textual data across diverse crisis contexts.

Abstract

Examining emotion interactions as an emotion network in social media offers key insights into human psychology, yet few studies have explored how fluctuations in such emotion network evolve during crises and normal times. This study proposes a novel computational approach grounded in network theory, leveraging large-scale Japanese social media data spanning varied crisis events (earthquakes and COVID-19 vaccination) and non-crisis periods over the past decade. Our analysis identifies and evaluates links between emotions through the co-occurrence of emotion-related concepts (words), revealing a stable structure of emotion network across situations and over time at the population level. We find that some emotion links (represented as link strength) such as emotion links associated with Tension are significantly strengthened during earthquake and pre-vaccination periods. However, the rank of emotion links remains highly intact. These findings challenge the assumption that emotion co-occurrence is context-based and offer a deeper understanding of emotions' intrinsic structure. Moreover, our network-based framework offers a systematic, scalable method for analyzing emotion co-occurrence dynamics, opening new avenues for psychological research using large-scale textual data.

Figures (4)

• Figure 1: Conceptual illustration of steps to build an emotional co-occurrence network based on co-occurring POMS concepts within a single post. (a) Step 1: Extract pairs of POMS concepts that co-exist within the same post. (b) Step 2: Construct a POMS concept network. After aggregating the co-occurring pairs of POMS concepts (words) over multiple posts, we get a network where each node represents an emotion concept, and each link strength represents the significance of co-occurrence between emotion concepts (see Methods). (c) Step 3: Build an emotion network by aggregating the number of significant concept links between and within emotion nodes.
• Figure 2: Stability of emotion and concept network over time (vaccination dataset). (a) Heatmap of Spearman rank correlation of emotion link's strength between emotion network snapshots. The part BELOW the diagonal shows the similarity of emotion networks including both intra- and inter-emotion links, while the part ABOVE the diagonal shows that of only inter-emotion links. The annotations indicate Spearman's rank correlation coefficient, along with the significance asterisks which depict the level of significance based on the $p$-value. The $p$-values are evaluated by the FDR. (b) Heatmap of Jaccard Index between temporal concept networks. The annotations indicate the value of Jaccard Index. In all sub-figures, we adopt thresholds of link weight $\ge$ top 10% and link strength $\ge$ 3 (see Methods). The color bar on the right of each subplot shows the mapping between colors and the corresponding data values, ranging from 0 to 1. The red dotted lines represent the commencement of vaccination in Japan in April 2021.
• Figure 3: Similarity of emotion and concept networks across datasets. (a) Heatmap of Spearman rank correlation of emotion link's strength between emotion networks of different datasets. The part BELOW the diagonal shows the similarity of emotion networks including both intra- and inter- emotion links, while the part ABOVE the diagonal shows that of only inter-emotion links. The annotations indicate the average Spearman's rank correlation coefficient, along with the significance asterisks which depict the level of significance based on the average $p$-value. The $p$-values are adjusted by the FDR. (b) Heatmap of Jaccard Index between concept networks of different datasets. The annotations indicate the average Jaccard Index. In all sub-figures, we adopt thresholds of link weight $\ge$ top 10% and link strength $\ge$ 3 (see Methods). The color bar on the right of each subplot shows the mapping between colors and the corresponding data values, ranging from 0 to 1.
• Figure 4: Comparing between the strength of emotion links in different datasets. (a) Visualization of emotion link strengths and the two-sample t-test across datasets. The strength of emotion links is rescaled so that the median of the strengths of each dataset becomes 1. The line charts show the median strength of inter- and intra-emotion links with 75% confidence interval for each dataset. The inset figure zooms into the logarithmic strength of inter-emotion links to better visualize the comparison between different datasets. The heatmap shows the statistics of two-sample t-tests to compare the distribution of emotion link strength between different datasets, with red color representing an increment of strength and blue color representing a reduction of strength. The asterisks show the significance of such changes, measured by the $p$-value of the $t$-test - *: $p \leq$ .05, **: $p \leq$ .01, ***: $p \leq$ .001. (b) Visualization of inter-emotion networks. The node represents emotion dimensions, thickness of links depicts the rescaled strength of link between two emotions. Abbreviation of datasets: EQ2011(All) - Earthquake2011 (all posts), EQ2011- Earthquake2011 (filtered by earthquake-related keywords), EQ2018(All) - Earthquake2018 (all posts), EQ2018- Earthquake2018 (filtered by earthquake-related keywords), BS: Bluesky.