Global Sentiment

Our Projects

Global Sentiment during COVID-19

How did the spread of the COVID-19 pandemic
affect well-being as measured on social media?

WHAT IS THE IMPACT OF THE PANDEMIC ON SENTIMENT? We measure and model sentiment measured on social media from January to June 2020.
HOW MENTALLY RESILIENT WERE DIFFERENT COUNTRIES DURING COVID-19? We quantify the magnitude of the sentiment shock, model the sentiment recovery dynamic, and examine global heterogeneities.
WHAT WAS THE IMPACT OF POLICY INTERVENTIONS? We evaluate the costs and benefits of different interventions, and the implications for future policy-making.

Global Sentiment and
Climate Change

What is the emotional toll of climate change?

What is the impact of climate events, such as temperature increases and environmental disasters, on subjective well-being and happiness? This project aims at addressing this issue by posing the following research questions:

What is the effect of climate change and environmental disasters on subjective psychology? We consider the link between the effects of climate change and subjective well-being. Using social media, we create a global sentiment index which we match to historical climate and environmental data. Controlling for seasonality and news cycles, we are able to estimate the sentiment change due to warming temperatures, increased weather unpredictability, and more frequent environmental disasters.
How has belief in Climate Change evolved over time? We also study how Climate Change is discussed on social media: in particular, when do people acknowledge the reality of Climate Change and the importance to do something about it? Among the phenomena we examine are: (1) when Climate Change is discussed seriously vs. jokingly; (2) which subtopics are associated with serious Climate Change-related conversations; (3) based on sentiment and key terms, how impactful (likes, retweets) is Climate Change-related content. In addition to providing us with an evolution of the way Climate Change is discussed on social media, we can evaluate the most effective communication on the subject by comparing sentiment and reach to the key terms used in the tweet.

PRELIMINARY RESULTS COMING SOON!

Climate Change in Portugal

How social media analysis can inform effective policy-making

Climate Change poses important risks in Portugal, including extreme temperatures, droughts, and rising sea levels. Insufficient research focuses on its perception and impact on wellbeing.
This project aims at understanding human attention and subjective evaluation through social media analysis. In conjecture with environmental data, these results can inform policy-makers on the “emotional toll of climate change” in Portugal.

SEE OUR PRELIMINARY FINDINGS

Wildfires and
Subjective Well-being
in Indonesia

Quantifying the Environmental and Health Impacts

Wildfires are an increasingly frequent problem in Indonesia—a result of both climate change and intensive agriculture.
This project uses our social media sentiment index to study the impact that fires have on the people of Indonesia. We collected 12.9 million tweets posted between July 31, 2019 and November 10, 2019, and compute the sentiment index. We also extract conversation topics and examine overall trends. Finally, we evaluate the impact of wildfires on subjective well-being, both directly—through topic modeling where we examine how often people discuss these fires—and indirectly—through an analysis of their expressed sentiment on social media, to understand people’s subjective well-being.
We find that both wildfires and air pollution have a negative impact on subjective well-being. These results establish an additional well-being cost to wildfires, and make their prevention and mitigation a clear policy priority.

LEARN MORE DOWNLOAD OUR FULL REPORT

Our Method

Our sentiment index relies on years of geolocated social media posts from Twitter and Weibo. Social media data is collected by the Harvard Center for Geographic Analysis Geotweet Archive and by the MIT Sustainable Urbanization Lab. Each post is encoded into vectorial representations using several Natural Language Processing (NLP) techniques, including dictionary-based methods (LIWC, Emoji dictionaries) and embedding-based algorithms (BERT). Pre-trained machine learning algorithms use these representations to impute the Twitter data’s sentiment and main topics, every day and everywhere.