[Paper Review] What a Nasty day: Exploring Mood-Weather Relationship from Twitter

Jiwei Li, Xun Wang+1 more|arXiv (Cornell University)|Oct 30, 2014

Climate Change Communication and PerceptionSocial Sciences44 references17 citations

TL;DR

This study investigates the mood-weather relationship using a two-year Twitter dataset and a sophisticated machine learning pipeline to identify explicit mood expressions while filtering out noise from public events and confounding factors. It confirms that high temperatures increase anger, snow correlates with depression, and overall weather conditions significantly influence multi-dimensional mood states, offering a scalable alternative to traditional psychological surveys.

ABSTRACT

While it has long been believed in psychology that weather somehow influences human's mood, the debates have been going on for decades about how they are correlated. In this paper, we try to study this long-lasting topic by harnessing a new source of data compared from traditional psychological researches: Twitter. We analyze 2 years' twitter data collected by twitter API which amounts to $10\%$ of all postings and try to reveal the correlations between multiple dimensional structure of human mood with meteorological effects. Some of our findings confirm existing hypotheses, while others contradict them. We are hopeful that our approach, along with the new data source, can shed on the long-going debates on weather-mood correlation.

Motivation & Objective

• To investigate the correlation between meteorological conditions and multi-dimensional human mood using large-scale, real-time social media data.
• To overcome limitations of traditional psychology surveys—such as small sample sizes, high cost, and susceptibility to individual biases—by leveraging Twitter's vast, continuous user-generated content.
• To develop a robust machine learning pipeline that identifies explicit mood expressions while filtering out non-mood-related factors like public events and linguistic ambiguity.
• To examine whether weather variables such as temperature, precipitation, and barometric pressure correlate with changes in collective mood, including anger, sadness, and happiness.
• To contribute a scalable, data-driven approach to the long-standing debate on weather-mood relationships, offering empirical insights beyond anecdotal or survey-based findings.

Proposed method

• Utilized a two-year Twitter dataset (10% of all tweets) collected via the Twitter API, focusing on geo-tagged messages to link mood with local weather conditions.
• Developed a machine learning pipeline to detect explicit mood expressions, distinguishing genuine emotional states from conventional phrases or negations (e.g., 'not happy').
• Applied sentiment and mood classification models trained to identify multi-dimensional affective states, including anger, sadness, and happiness.
• Implemented statistical controls to mitigate confounding factors such as public events (e.g., Michael Jackson’s death, Haiti earthquake) that could distort mood signals.
• Used regression models with interaction terms to address multicollinearity among weather variables (e.g., temperature, humidity, barometric pressure) and improve model robustness.
• Evaluated model performance using deviance explained (44.2%) and compared against a mixture model baseline (30.8%), confirming improved fit and reliability.

Experimental results

Research questions

• RQ1 Does high temperature correlate with increased expressions of anger in Twitter users?
• RQ2 Does snowfall or cold weather correlate with higher levels of sadness or depressive sentiment in social media posts?
• RQ3 Are there significant correlations between specific weather conditions (e.g., sunshine, barometric pressure) and positive or negative affect in collective mood?
• RQ4 To what extent can public events or non-weather-related factors confound the observed mood-weather relationships in social media data?
• RQ5 Can a machine learning pipeline effectively isolate genuine mood expressions from linguistic noise and conventional phrases in Twitter data?

Key findings

• High temperatures are significantly associated with increased expressions of anger, confirming long-standing hypotheses about heat-related aggression.
• Snowfall is correlated with higher levels of sadness and depressive sentiment, supporting the link between winter weather and seasonal affective disorder.
• The model explains 44.2% of deviance in mood variation, outperforming a baseline mixture model (30.8%), indicating strong statistical fit and reliability.
• Despite challenges from confounding factors like air pollution and event-driven mood shifts, the aggregated analysis successfully isolates weather-related mood patterns.
• The study confirms that weather variables such as temperature, precipitation, and barometric pressure have measurable, multi-dimensional effects on collective mood as expressed on Twitter.
• Contradictory findings from traditional psychology studies are partially reconciled by the large-scale, real-time nature of Twitter data, which reduces individual bias and increases statistical power.