Quantifying Collective Emotions: Japan's Societal Trends Through Enhanced Sentiment Index Using POMS2 and SNS

TL;DR

This paper develops a real-time seven-dimensional sentiment index for Japanese society by applying POMS2 categories plus a newly introduced Friendliness indicator to X post data. Indicators are computed from a dictionary-based word-count pipeline and normalized using a generalized average with $\alpha = 0.5$, producing time series $I_e(t)$ for each emotion. The authors use Prophet to decompose these series into long-term trends and seasonal components and introduce $\Delta I_e(t) = (I_e(t) - \hat{I_e}(t)) / \hat{I_e}(t)$ to quantify event-specific impacts, providing a framework to compare diverse societal events. The results show the index captures typical emotional fluctuations and echoes prior blog-based findings, while the Friendliness dimension adds a new metric of collective calmness and cultural responsiveness. The approach demonstrates robustness across data sources and holds practical value for visualizing societal trends and informing policy analytics, with future work on dictionary automation and linking events to external information, including adoption by the Nomura Research Institute as the NRI Sentiment Index.

Abstract

In this study, we constructed an emotion index that quantitatively represents the collective emotions present in the Japanese web space by utilizing Social Networking Service (SNS) post data. Building upon previous research that used blog data and the Profile of Mood States (POMS), we restructured the methodology using posts from X (formerly Twitter) and updated the model by adding the ``Friendliness" indicator from the POMS2 metrics. Through periodic and trend analyses of the emotional indicators derived from X's post data, we found that the extension is consistent with results previously reported using blog data. This suggests that our methodology effectively captures typical emotional fluctuations in Japanese society, independent of specific SNS platforms, and is expected to serve as an index to visualize societal trends.

Figures (7)

• Figure 1: POMS-based sentiment index proposed by Sano, Fig.1(A) in POMS_PLOS
• Figure 2: Periodical properties of the sentiment index, Fig.2(A) in POMS_PLOS
• Figure 3: Schematic view of the calculation flow of the sentiment index
• Figure 4: the contribution ratio of the post counts for words defined by the "Anger" indicator to the indicator value. The left shows the ratio of post counts to the total is averaged to construct the indicator. The right is the ratio using the generalized average with $\alpha =0.5$.
• Figure 5: The indicator values for each emotion in the Japan Emotion Index defined in this study. The values are normalized so that the period from 2021 to 2023 is set to 1. Dotted lines link relevant events to the points where the time series shows significant fluctuations (peaks).