Users on social media sites like Twitter are increasingly relying on crowdsourced recommendations called Trending Topics to find important events and breaking news stories. Topics (mostly keywords, e.g., hashtags) are recommended as trending when they exhibit a sharp spike in their popularity, i.e., their usages by the crowds suddenly jump at a particular time.
While prior works have attempted to classify and predict Twitter trending topics, in this work, we ask a different question — who are the users who make different topics worthy of being recommended as trending?
Specifically, we analyse the demographics of the crowds promoting different trending topics on the Twitter social media. By promoters of a topic, we refer to the users who posted on the topic before it became trending, thereby contributing to the topic’s selection as a trend.
We gathered extensive data from Twitter from July to September, 2016, including millions of users posting on thousands of topics, both before and after the topics became trending. We inferred three demographic attributes for these Twitter users — their gender, race (Asian / Black / White), and age — from their profile photos.
Looking at the demographics of the promoters reveals interesting patterns. For instance, here are the gender and racial demographics of the promoters of some of the Twitter trends on 3rd May 2017:
- #wikileaks: 24% women, and 76% men
- #wednesdayWisdom: 52% women, and 48% men
- #comey: 9% Asian, 12% Black, and 79% White
- #BlackWomenAtWork: 15% Asian, 52% Black, and 33% White
It is evident that different trends are promoted by widely different demographic groups.
Our analysis led to the following insights:
- A large fraction of trending topics are promoted by crowds whose demographics are significantly different from Twitter’s overall user population.
- We find clear evidence of under-representation of certain demographic groups among the promoters of trending topics, with mid-aged-black-females being the most under-represented group.
- Once a topic becomes trending, it is adopted (i.e., posted) by users whose demographics are less divergent from the overall Twitter population, compared to the users who were promoting the topic before it became trending.
- Topics promoted predominantly by a single demographic group tend to be of niche interest to that particular group.
- During events of wider interest (e.g., national elections, police shootings), the topics promoted by different demographic groups tend to reflect their diverse perspectives, which could help understand the different facets of public opinion.
Try out our Twitter app to check demographics of the crowds promoting various trends.
For details, see our full paper, Who Makes Trends? Understanding Demographic Biases in Crowdsourced Recommendations, at ICWSM 2017.
Abhijnan Chakraborty, IIT Kharagpur, India and MPI-SWS, Germany
Johnnatan Messias, Federal University of Minas Gerais, Brazil
Fabricio Benevenuto, Federal University of Minas Gerais, Brazil
Saptarshi Ghosh, IIT Kharagpur, India
Niloy Ganguly, IIT Kharagpur, India
Krishna P. Gummadi, MPI-SWS, Germany