Before connecting with Moore, Obradovich had been using the streaming endpoints on the Twitter API to collect data over the years to understand how people’s sentiment changed over time when exposed to different weather. “It was very serendipitous [when Moore reached out with her research question], we had already created classifiers for parsing weather-related Tweets from non-weather related Tweets, and had worked on understanding the sentiment of these Tweets,” said Obradovich. “We had validated these data in varying settings, even across different social platforms. So when [Moore] reached out, we realized we could map our approach differently over to our dataset of Tweets to answer the research questions [Moore] had been pondering.”
Together, Moore and Obradovich spent two years evaluating over 2 billion Tweets that were geolocated in the continental United States, pulling out Tweets related to weather. They measured peoples’ sentiment as well as the ‘remarkability’ of different temperatures (the frequency of talking about weather) and how it changes with repeated exposure to unusual temperatures.
In addition to the mapping these classifiers over the Tweet dataset to perform sentiment analysis, the team went on to merge in climate/weather data, and aggregate Twitter data to different spatial levels – such as by county or major cities that have seen extreme weather events.
Leveraging Twitter data sourced from streaming endpoints on the Twitter API, they were able to answer the boiling frog question and published their findings, which have been widely received.