Measuring the Changes in Sentiment and Emotion Towards COVID-19 Over Time in Tweets Posted from Within United States Counties

All mentors and contributors: Zachary Richardson, James Stassinos, Taylor Anderson, Amira Roess, Hamdi Kavak, Andreas Züfle The COVID-19 pandemic was a difficult time for the world and many people took to Twitter to discuss their thoughts and feelings towards COVID-19. Understanding the sentiment and emotions felt by a population can give insight on how the general public’s opinions are changing as the pandemic progresses. We can see trends in sentiment and discover if online behavior is an indicator of offline behavior in response to COVID-19. To begin our research, we used a dataset created by the Qatar Computing Research Institute (QCRI) that contains millions of tweets with enriched geolocation information attributed to each tweet. We first had to extract tweets posted within Fairfax County, then perform our own sentiment and emotion analysis. To calculate sentiment, we used the python library TextBlob to read the text within tweets and give it a score within the range -1 to 1 based on how negative or positive the attitude of text was. To measure emotion, we used the python library Text2Emotion to give tweets a percentage score based on the five emotions: happy, sad, angry, surprise, and fear. We aggregated the sentiment scores and emotion percentages by day from February 1st, 2020, to May 1st , 2020. We found that on average fear was the top emotion felt by Twitter users within Fairfax County, and the sentiment was fairly neutral. It is difficult to see any trends within Fairfax County alone, our future work will extend to every county within the United States.

00:04 hi my name is zachary richardson 00:06 and my name is james stasinos over the 00:08 summer james and i have been working on 00:10 our project 00:10 measuring changes in sentiment and 00:12 emotion towards coven 19 00:14 over time in tweets posted within united 00:16 states counties 00:20 the coven 19 pandemic was an 00:21 unprecedented time for the world and 00:23 caused many to develop strong feelings 00:25 towards the situation 00:26 many in the united states took to 00:28 twitter to discuss and share their 00:29 thoughts and emotions towards copenhagen 00:32 in our project we aimed to measure these 00:34 emotions and overall sentiment of kobe 00:36 19 00:36 through these tweets shared by users 00:38 within the united states 00:39 as the pandemic progressed to begin this 00:42 research we obtained a data set put 00:44 together by the qatar 00:45 computing research institute the data 00:48 set contained hundreds of millions of 00:50 dehydrated tweet ids 00:51 with enriched geolocation information 00:54 they had all been 00:56 the enriched geolocation information had 00:58 been obtained 00:59 using advanced machine learning 01:01 techniques to make predictions about 01:03 user location 01:05 based on information within the profile 01:09 [Music] 01:10 all the tweets were found by searching 01:12 for tweets mentioning coven 19 or covet 01:15 19 related keywords for example social 01:18 distancing or quarantine 01:23 we use the developer account to access 01:25 the twitter api 01:26 to hydrate the tweets obtained through 01:28 the qatar dataset 01:30 using the enriched geolocation 01:32 information we identified all the tweets 01:34 from the users located within fairfax 01:36 county 01:37 and we calculated the sentiment and the 01:39 emotion of those tweets 01:44 to measure the overall sentiment of the 01:46 tweets we use the python library called 01:48 textblob a natural language processor 01:51 that evaluates text data to determine 01:52 the tweet’s attitude 01:54 we use this library when reading the 01:56 text within the tweets and each tweet is 01:58 given a score within the range 01:59 negative one to one based on how 02:02 negative positive 02:03 or neutral the text was we aggregated 02:05 the average sentiment of every tweet 02:07 collected within fairfax county by day 02:09 within the time period of february 1st 02:11 2020 02:12 to may 1st 2020 and we discovered the 02:15 overall sentiment was fairly neutral at 02:17 the beginning of the pandemic 02:21 to measure a motion of the tweets we use 02:23 the python library called text 02:24 to emotion another natural language 02:27 processor that allows us to extract 02:29 five emotions from the text data those 02:31 emotions are happiness 02:32 anger sadness surprise and fear 02:35 each emotion is assigned a percentage 02:37 based on how much 02:38 that text corresponds to each of the 02:40 five emotions 02:42 we aggregated the average emotion score 02:44 received from each tweet collected 02:46 within fairfax county by day within the 02:48 time period of february 1st 02:50 2020 to may 1st 2020 and we discovered 02:52 the top emotion built by twitter users 02:54 in the early stages of the pandemic was 02:56 fear 02:57 followed closely by sadness and surprise 03:02 we are currently working on creating a 03:04 data set of sentiment and emotion for 03:06 all the covet 19 related tweets where 03:08 the user is located within the united 03:10 states 03:12 in addition the qatar computing research 03:14 institute team 03:15 will be publishing an updated covet 19 03:17 data set including more keywords 03:20 over a longer period of time during the 03:22 pandemic 03:24 which will include additional tweets 03:26 collected 03:28 we look forward to processing all that 03:30 additional data for sentiment and 03:31 emotion 03:32 to compare the different locations 03:34 sentiment and emotion 03:35 to learn about their different reactions 03:37 to the pandemic 03:38 as it forget as it progressed 03:44 thank you very much for viewing our 03:45 project uh we’d like to take this time 03:47 to acknowledge george mason university 03:49 office of provist 03:50 and the summer team impact project for 03:53 allowing us to conduct our research 03:54 uh we’re excited to see more results in 03:57 the future 03:58 thanks thank you