How do Twitter users use the term “resilience” in tweets related to different types of socio-economic shocks?

People use the social media platform Twitter to communicate information ranging from the quality of their latest meal to their pleas for emergency assistance during life-threatening hurricanes and tornadoes. The massive volume of tweets and the diversity of topics people discuss on the platform has made analyzing Twitter data a popular activity for academic researchers, media outlets, advertising agencies, and other groups interested in gauging public opinion. In this project we are using Twitter data to gauge the pulse of public opinion regarding the terms “resilience” and “resilient.”

We will use machine learning and topic modeling techniques to determine how these words are used in everyday Twitter dialogues, the topics people use the words to discuss, and the sentiment of the tweets in which the words are used. For example, we will measure the frequency with which Twitter users use “resilience” to discuss existential shocks, like the COVID-19 pandemic, versus their use of the word in discussions of less serious topics. We will determine which of these resilience topics are discussed in the most positive (or negative) language. Since this project is designed as a longitudinal study, we will be able to track and measure how “resilience” topics, sentiments, and dialogues change over time in response to different events. Since July 2020, we have collected more than 3 million tweets that include the words “resilience” and/or “resilient” which indicates that the terms are widely used on Twitter and merit the type of investigation we are conducting.

By understanding how “resilience” is used on social media we will gain valuable insights that will enable us to compare academic literature regarding the concept with the way the larger public conceptualizes and uses the term. This new understanding can help facilitate the dialogue between academic researchers who study resilience and the general public regarding a wide variety of challenges and shocks that affect economic outcomes, community health, and many other topics that shape the way we live our lives.

• What are the topics people use the terms “resilience” and “resilient” to discuss on Twitter?
• Do Twitter users use these terms to discuss economic downturns, natural disasters, pandemics, and other shocks, or are the terms used to discuss other topics?
• To what extent do the “resilience” topics discussed on Twitter compare to those studied by academics?
• Considering the different topics, what is the sentiment (positive/negative) of the average resilience tweet? Are certain topics discussed with more positive or negative sentiment than others?
• Which words are used with the greatest frequency in the tweets associated with each “resilience” topic?
• How do the topics and sentiment of tweets containing certain words change over time? For example, how do the tweets with the words “resilience” and “COVID-19” change over time?
• How do the topics of tweets with certain words change over time in comparison to other data sources? For example, do the topics and sentiments of tweets with the words “resilience” and “COVID-19” change in ways that reflect increasing morbidity and mortality rates over time? Does the sentiment of tweets become more negative as rates rise?

This project is a partnership between the Knowledge Exchange for Resilience and the Decision Theater®.

• Using the “Pulse” tool developed by ASU’s Decision Theater®, we collect and store tweets in a database for analysis. All tweets containing the words “resilience” and/or “resilient” are added to the database in real time, whether the word appears within the normal text of the tweet or as a hashtag. Currently we’ve accumulated over 3 million tweets, with an average of over 27,600 tweets per day. The Pulse tool also provides a dashboard for some analysis including filtering the data set further by certain words, and visualizing time series peaks, most popular hashtags, and social network webs of the most prominent users or mentions within the collected tweets. Additionally, the database can be exported to a .csv file for further analysis.

• We will use machine learning techniques and topic modeling techniques such as Latent Dirichlet Allocation (LDA) to identify the resilience topics in the Twitter data.

• By using these machine learning and topic modeling methods in a longitudinal research design, we will be able to identify the resilience topics discussed by Twitter users and monitor how they change over time.

• We will use sentiment analysis techniques to measure the positivity/negativity of the tweets.

• We will create dictionaries of the words used most frequently to discuss the resilience topics we identify using the topic modeling techniques.

• We will determine if there are influential Twitter users involved in the different resilience topic discussions by counting the number of resilience tweets per Twitter user and the number of retweets.

• By doing this analysis, we can not only create a better understanding regarding the similarities and differences between the way resilience is discussed on Twitter versus in academic research. Please see the project titled “Social networks, social capital, spatial analysis and the resilience puzzle: An exploration of the current use of theories, methods, and metrics in resilience research” for more information regarding how we are analyzing the resilience topics studied by academic researchers.

• This research can help us focus our resilience research design efforts in the future.
• This research is an example of ASU’s Transform Society and Engage Globally design aspirations. In the Transform Society aspiration, ASU catalyzes social change by being connected to social needs. In this research, we are analyzing social media data that provides people with a platform to express a wide variety of opinions, thoughts, feelings, and social needs. This research Engages Globally because Twitter is accessible as a communication platform for people at local, national, and international scales to discuss issues and express thoughts and opinions. With Twitter data, we are able to analyze people’s thoughts and opinions regarding resilience topics across all three scales.
• By identifying the topics discussed on Twitter, we can get a better understanding regarding which of Rodin’s principles social media users discuss, how they discuss them, and how their discussions change over time in response to different types of shocks and current events. By doing this, we will learn something regarding public discussions regarding resilience and how they compare to Rodin’s principles.