Friday, October 20, 2017

What Tweets Get the Most Engagement on Trending Topics? A Case Study of #MeToo by Mathew L

Note: This blog entry mentions the topics of sexual harassment and assault.
Background:
            Over the last week, #MeToo has been trending on Twitter after a call to action from actress Alyssa Milano for people to highlight their experiences of sexual harassment or assault and show “the magnitude of the problem”.[1] The high volume of #MeToo tweets can put to the test theories of how trends develop over time on Twitter, providing an empirical basis to a field that currently has few theories and even fewer good case studies. One area I will examine in this project is how #MeToo manifests itself over Twitter as the days go by past its initial peak popularity.  I am also interested in examining what types of Tweets in #MeToo were able to rise to the top in terms of engagement, and whether there was a shift in their type over time too. I hope to be able to produce some observations about what type of Tweet drives trending social movements on Twitter to provide insight on how to best use social media to create discourse or spread ideas. 
Questions to answer:
1.     How did #MeToo evolve over time on Twitter as a trend?
2.     What type of Tweet drew the most engagement in #MeToo every day?
Hypothesis:
1. Tweets on #MeToo become more about social commentary and less of solidarity/sharing stories
Drawing intuitions purely from news media coverage of the social trend, the hashtag seemed to initially be limited to a focus on Harvey Weinstein and issues of sexual assault and harassment in the U.S. entertainment industry.[2] This initial reason for the hashtag to trend was then usurped by the broader response that spanned way beyond Hollywood as many more users shared their own experiences with sexual assault and harassment. Currently, a broader discussion also has sprouted out on cultural issues that created the enabling environment for the incidents to happen. Over the next week or so we will probably see a decrease in the number of shared stories and the hashtag will reflect more instances of social commentary. Eventually the hashtag will see momentary resurgences when similar news events about sexual assault and harassment capture the attention of the media, much like how Black Lives Matter as a slogan tends to waver in terms of trends according to whether there are further accounts of police brutality. A visualization of the total tweets of #MeToo per day over time may look like an inverted U Shape that eventually tapers off into smaller oscillations as the news cycle captures new stories of sexual assault or harassment.

2. Tweets with a “call to action” draw more engagement
While the number of engagements a tweet receives will definitely depend on the number of followers the account already has, the content of the Tweet will also be a significant determinant. Tweets that suggest a course of action, such as for men to speak out or for more resources to report sexual misconduct will appear to add more facets to the discussion and draw a bigger response. Retweets of #MeToo, while critical in contributing to the movement over time will draw less attention than other tweets that provide additional input. While a compelling personal story may also draw lots of response too, it may be harder to shape a powerful story in just 140 letters compared to calls to action so the calls to action Tweets will in aggregate draw more engagement.
Methodology:
            Data collection will be assisted by NodeXL, using its limit of drawing 18,000 tweets per day that used the hashtag on #MeToo. A snapshot of tweets will be taken once every day, as this is the full capacity of Tweets that can be mined. With this dataset of Tweets the attributes of location, user’s number of followers, user’s number of tweets, retweets of the tweet and favorites of the tweet will be indexed.
            Using the attribute data, I will extract the top 20 and bottom 20 Tweets in terms of total retweets and favorites from each day as Tweets of interest. I will index the Tweets according to what type of response they were communicating with #MeToo. On a broader level this approach has been applied by by Zubiaga et. that created a typology on trending topics on Twitter to be categorized into News, Ongoing Events, Memes, and Commemoratives.[3]  This approach of categorizing trends will be extended to illustrate what types of engagement with the topic drew the most and least attention. After browsing over the raw data I currently have these are the 8 types of categories I intend to sort by:
1. Retweets only of #MeToo 
2. Critiques or social commentary 
3. Longer-formed personal narrative about own experiences 
4. Expressions of remorse or guilt from men 
5. Doubts, backlash, and negative responses
 6. Humorous responses 
7. Corporate/verified accounts making statements 
8. Media outlet coverage. 

              In addition to this sorting, I will have a binary index of whether these top Tweets included a “call to action” in them. A final level of analysis I will bring in is to identify if the top tweets were responses to a longer thread and how long of a thread were they a part of. This part is where the visualization of social network analysis may be very helpful as top responses can all be a part of the same thread of replies. The network I will be looking at thus includes just the selected Tweets with most and least engagement. While the top engagement Tweets are more of interest, the least engaging ones will also be analyzed in the same methodology to potentially strengthen the confidence of any observations. The most important network measure of all will be the attribute data of what type of response the Tweets were under my categorization.
            With these methods, I will be focusing on whether the types of responses that get the most engagement change over time. This will directly test the hypothesis that the type of top engagement shifts more towards an activism/social critique direction over time beyond just sharing of personal experiences. Furthermore, the proportion of tweets that include a call to action will test the second hypothesis that prompting action from others may resonate more in terms of garnering more Twitter engagements.
Conclusion/Challenges and Limitations:
            The limitations of this methodology will be Tweets that fall through the cracks, missing the top trending Tweets that do not appear during the one data collection point per day. Since the 18,000 Tweets are only the most recently available ones, they cannot track the entire activity around #MeToo as its total Tweets currently vastly exceed 18,000 a day. However, it may be interesting to see if over the course of the next few weeks the daily number of Tweets eventually fall below 18,000. A further limitation is the identifying information like location or type of person handling each account cannot be verified for accuracy unless they have a blue check mark. One may describe themselves as “a student from Paris” but it is pretty hard to take this information at face value. I will be sure to keep this in mind while making qualitative observations.  All in all, I hope this project will produce some viable takeaways on what type of activity on Twitter receives the most engagement on trending topics.

Additional Sources:

Derek L. Hansen, Ben Shneiderman and Marc A. Smith, Chapter 2 - Social Media: New Technologies of Collaboration, In Analyzing Social Media Networks with NodeXL, Morgan Kaufmann, Boston, 2011, Pages 11-29, ISBN 9780123822291, https://doi.org/10.1016/B978-0-12-382229-1.00002-3.
(https://www.sciencedirect.com/science/article/pii/B9780123822291000023)

Kawash, Jalal. Online Social Media Analysis and Visualization. Lecture Notes in Social Networks (Unnumbered). 2014.




[2] Ibid.
[3] Zubiaga, A., Spina, D., Martínez, R. and Fresno, V. (2015), Real-time classification of Twitter trends. J Assn Inf Sci Tec, 66: 462–473. doi:10.1002/asi.23186

1 comment:

Christopher Tunnard said...

Nice work, Mathew, and the beginnings of a rewarding project. You point out the benefits and shortcomings of doing this analysis clearly, and I think you've got realistic expectations of the types of conclusions you can draw. As we discussed in class, you need to work on your Key Question; doing that will help you manage the scope.