Election chat: the secrets of youth social media revealed

A new social media mining study conducted by Well Told Story and Africa’s Voices Foundation (AVF) shows that Shujaaz fans are on average 2-3 times as engaged in conversations about governance as in conversations on other topics on DJ B’s Facebook page.

In preparation for the 2017 Presidential Elections in Kenya, AVF and WTS joined forces to explore social media discourse around governance. The study used public data from DJ B’s Facebook page to explore the lifecycle of digital conversations and factors that can affect such conversation.

Social media data gives researchers unprecedented access to digital conversations at a massive scale. DJ B is Shujaaz’s lead character & his Facebook page is a perfect example of social media’s Big Data. Since its launch, DJ B’s page has been liked by 473,014 people. In the past 12 months, DJ B has made 6,492 posts, which have inspired 106,089 comments (from 37,835 unique users) and 653,427 reactions (from 176,263 unique users).

This dynamic on-going conversation between DJ B and his fans inspired the AVF-WTS team to conduct a study to explore how youth talk online. For example, what topics do fans prefer? Why does certain content go “viral” while others not? What is the prevalent ‘mood’ or sentiment around various themes?

For the first round of analysis we selected a subset of 157 posts that focused on the issues of governance published between October 2016 and July 2017.  These posts inspired 6,955 comments (by 4854 unique users) and 53,960 reactions (by 34,273 unique users). When we compare the engagement created by these posts with all conversations led by DJ B (as per Table 1), we can see that governance posts enjoy substantially higher levels of fan engagement vs. other topics on DJ B’s page.

Table 1: The historic average of comments, likes and shares around the governance posts by DJ B vs. posts on other topics.

In addition, the observed fan activities (comments, likes and shares) have been significantly growing in the run-up to the elections (between April and July 2017, Figure 1)

Figure 1: The monthly average of comments, likes and shares around the governance posts by DJ Boyie, April – July 2017.

In the second round of analysis, we explored the intensity and polarity of fans engagement around the four most popular posts on governance using the same analytical framework we developed for contraception conversation analysis (discussed here). Our approach included (1) collaborating with trained WTS annotators (who speak English, Swahili and Sheng) to label the sentiment in the comments as positive, negative, or neutral; and (2) using regular expressions and a custom lexicon to isolate the topics found in textual data. This dual approach allowed us to segment and classify comments based on their topic and sentiment, analyze the conversations, and prepare for further analysis using machine learning.

The first selected post asked fans which issue affected them more – corruption or unemployment?

This post inspired 300 comments and 1400 reactions. Of those, 227 comments (76%) chose corruption as the key issue, 49 (16%) opted for unemployment, and 22 comments (6%) mentioned other areas such as healthcare.

The concern about corruption was replicated in the second most popular post (below), which broadened the conversation to include such issues as high prices, poverty, security/crime, drought/famine and healthcare.

This second post inspired 91 comments. Corruption was brought up in 36 comments, unemployment in 19 comments, poverty in 15 comments, drought/famine, security/crime, and high prices each in 7 comments and healthcare in 1 comment.

Overall, corruption and unemployment were the two most common topics, recurring throughout the entire 10-month conversation on governance inspired by the 157 posts by DJ B.

In the last round of analysis, we tested the sentiment classifier, i.e., comment labeling as negative, neutral or positive.  For the testing, we used two posts. The first one asked Shujaaz fans to comment about their experiences with the police.

This post inspired over 2,100 reactions and 270 comments. When we isolated irrelevant comments (n=31), among the remaining 239 comments 88 were classified as positive, 60 as neutral and 60 were negative.

In the second post used for sentiment-classifier testing, DJ B asked fans about their views on the new devolved county system in Kenya (which, many argue, will play a significant role in the upcoming elections).This post inspired 63 likes and 33 comments of which 62% of comments reported positive experiences about the county system and 38% expressed negative sentiment.

In the weeks and months after the elections we will continue to explore the content of Shujaaz fans’ conversations and, importantly, how this content relates to social networks behind it: who are the influencers among Shujaaz fans? How do Shujaaz messages travel through secondary networks of Shujaaz fans in other social media communities? etc. This hybrid analysis will not only help us better understand how people engage with relevant content around governance but also how different types of content find resonance across broader social media communities and, finally, may spill over to social norms and practices in the offline world.

Written by: Dr. Matti Pohjonen, Senior Researcher & Elisabeth Kerr, Researcher – Africa’s Voices Foundation

AFRICA’S VOICES FOUNDATION AND WELL TOLD STORY

Since 2014, Africa’s Voices Foundation has been working closely together with Well Told Story to understand digital conversations of the Shujaaz media audience. This collaboration has led to exciting new insights about how digital conversations reflect normative changes and shifts in behaviors among Kenyan youth (15-24 yo); it also enabled the discovery of new research approaches, including the description of the anatomy of sentiment and the method to research low-resource languages such as Sheng. You can read more about other joint projects between AVF and WTS here and here.