Who Makes Trends? Understanding Demographic Biases in Crowdsourced Recommendations

Abhijnan Chakraborty, Johnnatan Messias, Fabrício Benevenuto, Saptarshi Ghosh, Niloy Ganguly, Krishna P. Gummadi

Introduction

Social media platforms increasingly rely on crowdsourced recommendation systems, such as Twitter’s "Trending Topics," to curate and highlight significant information. These systems serve as powerful gatekeepers; once a topic trends, it achieves massive visibility and often crosses over into traditional news media—a process known as "hashtag journalism." However, this mechanism assumes that the "crowd" is a representative body. This study investigates the demographic makeup of the users who drive these trends, revealing that the individuals responsible for promoting topics before they go viral are often highly unrepresentative of the broader platform population.

Methodology

To quantify demographic bias in trend promotion, the researchers developed a multi-stage pipeline:

  1. Data Collection: Monitoring US trending topics every five minutes over three months in 2016, alongside a 1% sample of all tweets (340 million tweets from 50 million users).
  2. Demographic Inference: Using the deep-learning platform Face++, the authors inferred gender, race, and age from profile pictures for a baseline of 1.7 million US-based users.
  3. Role Definition: The study distinguished between Promoters (those posting before a topic trends) and Adopters (those joining after).
  4. Statistical Analysis: Fisher’s Exact Test and the "80% rule" (a US employment law standard) were used to identify systematic under-representation compared to the general Twitter population and US Census data.

Demographic Divergence in Trend Promotion

The study found that trends are rarely driven by a general cross-section of the public. Instead, they are typically the product of specific demographic clusters. For instance, gaming-related hashtags were heavily dominated by male promoters, while entertainment trends like #BacheloretteFinale were driven by women.

Racial demographic distribution of trend promoters across different hashtags, showing how trends often cluster by specific racial compositions.

Statistically, 88.25% of the analyzed trends showed promoter demographics that significantly diverged from the overall Twitter population across gender, race, and age. Interestingly, once a topic becomes a trend, the demographics of "adopters" tend to normalize, suggesting that while trending status pulls in a diverse audience, the initial "spark" is inherently biased.

The "Glass Ceiling" Effect

A primary finding of the research is that marginalized groups in society often face a "glass ceiling" in the trend-making process. Despite their active presence on the platform, certain groups are systematically under-represented in the promoter phase.

Percentage of trends where specific demographic groups are under-represented, highlighting that Black females are the most frequently marginalized group in trend promotion.

As shown in the data, women and Black users are frequently under-represented relative to their share of the Twitter population. Specifically, mid-aged Black females were identified as the most under-represented group among trend promoters. This suggests that the interests and perspectives of these groups are less likely to be elevated by the platform's current crowdsourcing algorithms.

Conclusion and Limitations

The research highlights a critical bias: "crowdsourced" recommendations often reflect "majority-sourced" or "dominant-group-sourced" interests. While the study is limited by the accuracy of AI-based demographic inference and the use of broad categorical labels (e.g., binary gender), it underscores the need for recommendation algorithms to explicitly account for demographic diversity. To increase transparency, the authors launched "Who-Makes-Trends," a web service designed to make these hidden biases visible to the public and platform designers alike.

Summary updated

Abstract

Users of social media sites like Facebook and Twitter rely on crowdsourced content recommendation systems (e.g., Trending Topics) to retrieve important and useful information. Contents selected for recommendation indirectly give the initial users who promoted (by liking or posting) the content an opportunity to propagate their messages to a wider audience. Hence, it is important to understand the demographics of people who make a content worthy of recommendation, and explore whether they are representative of the media site's overall population. In this work, using extensive data collected from Twitter, we make the first attempt to quantify and explore the demographic biases in the crowdsourced recommendations. Our analysis, focusing on the selection of trending topics, finds that a large fraction of trends are promoted by crowds whose demographics are significantly different from the overall Twitter population. More worryingly, we find that certain demographic groups are systematically under-represented among the promoters of the trending topics. To make the demographic biases in Twitter trends more transparent, we developed and deployed a Web-based service 'Who-Makes-Trends' at twitter-app.mpi-sws.org/who-makes-trends.
BibTeX
@inproceedings{chakraborty2017who,
  author = {Abhijnan Chakraborty and Johnnatan Messias and Fabrício Benevenuto and Saptarshi Ghosh and Niloy Ganguly and Krishna P. Gummadi},
  title = {{Who Makes Trends? Understanding Demographic Biases in Crowdsourced Recommendations}},
  booktitle = {ICWSM},
  year = {2017},
  url = {https://openreview.net/forum?id=EvS6CwAH7uc},
}
Figure