By pulling tweets from the Twitter Platform, researchers hoped to gauge the real-time sentiment of individuals on a wide scale. Although researchers were able to analyze the sentiment of Twitter Users through their tweets to obtain additional insight, researchers determined there was not enough evidence to suggest that the count of friends, count of followers, count of favorites, count of retweets, and result type of a given tweet could be predicted by the sentiment label of the tweet. However, it appeared that scores of Emotional/Physical Well-Being had potential to be a good predictor of overall Happiness Scores. Additionally, it was determined that there was a significantly different happiness scores and in count of friends on Twitter across the 50 U.S. states.
Considerations of Weaknesses of the Study
As with any study, there are some considerations as it pertains to the assumptions made. The conclusions drawn as part of this study were heavily reliant upon the accuracy of the sentiment assignment of content of the individual tweets. While we were able to confirm the sentiment was largely in-sync with hand-assigned sentiment, both humans and algorithms have difficulty understanding sarcasm in tweets due to the limited amount of context and text. With this in mind, if the sentiment scores assigned to the individual tweets were significantly off, there is a potential that the conclusions drawn in this study may be flawed.
Additionally, the findings relied heavily on the assumption that WalletHub’s Happiest States findings [3] were accurate. Because the state happiness rankings were used as a foundational baseline, any inaccuracies in this data could potentially impact the reliability of the analyses conducted in this study.
Exploration of Possible Directions and Goals for Future Analysis
Potentially, additional levels of analysis can be conducted on both the Happiest States Data and the Twitter Data to determine whether text or non-textual characters, including emoticons and other symbols, are better predictors of the overall happiness of a U.S. state, with a high level of precision. Clustering of states based on location, nearest neighbor, or latitude, could potentially be applied to determine wider impacts of weather or geographic area on the United States population happiness and tweet sentiment.
We can also take advantage of the live-feed aspect of Twitter to examine sentiment on a more microscopic level, analyzing how major events may impact the overall sentiment of tweets, across locations, on a given day. In wake of tragedies like the October 2017 mass shooting in Las Vegas, Nevada, is tweet sentiment greatly affected? Additionally, are certain state populations more susceptible to change in overall tweet sentiment following a given major current event?
Is there a differentiation between tweet sentiment based on day of the week? Can a significant difference in tweet sentiment be detected on Mondays as compared to Fridays, or on weekdays versus the weekends?
