Can A User Anticipate What Her Followers Want?

Abir De, Adish Singla, Utkarsh Upadhyay, Manuel Gomez-Rodriguez

Whenever a social media user shares a story, they are typically seeking feedback—likes, comments, or shares. This interaction creates a feedback loop where the user may feel compelled to re-estimate their followers' preferences based on past engagement to decide what to post next. This paper investigates the conditions under which a user can successfully anticipate their followers' desires and provides a framework to identify this behavior in the wild.

The Social Media Feedback Loop as a Decision Problem

The researchers from the Max Planck Institute model the social media experience as a sequential decision-making task. In this model, a user acts as a "utility maximizer" who balances two competing interests: their own topic preferences and the feedback they receive from their followers.

This problem is analogous to the "Multi-Armed Bandit" problem in reinforcement learning. Each time a user posts, they are essentially pulling a lever on a slot machine (a topic) to see what the payout (feedback) will be. To maximize their total utility over time, the user must navigate the exploration-exploitation trade-off:

  • Exploitation: Sharing stories and topics that have historically led to high engagement.
  • Exploration: Sharing new or diverse stories to learn more about the followers' evolving preferences.

Theoretical Insights: The Advantage of Social Learning

The paper provides a rigorous theoretical analysis of different posting strategies. A key finding is that users who rely on simple "point estimates"—essentially just looking at the average likes of past posts—often fail to optimize their strategy. These users get stuck in "linear regret," meaning they never truly learn the optimal content mix because they don't account for the uncertainty of what they haven't tried yet.

In contrast, the authors show that users who utilize Posterior Sampling (Thompson Sampling)—treating follower preferences as a probability distribution—perform significantly better, achieving logarithmic regret.

Perhaps the most striking insight is the power of Social Learning. If a user can observe the feedback their followers provide to other users, they can effectively skip the risky exploration phase. By learning from the broader ecosystem, a user can achieve "constant regret," reaching near-optimal engagement almost immediately without having to experiment with unpopular posts themselves.

Detecting Feedback-Sensitive Behavior

To move from theory to practice, the authors developed a utility estimation framework designed for observational data. This framework uses Maximum Likelihood Estimation and Statistical Hypothesis Testing to determine if a specific user is actually changing their behavior based on feedback or simply posting according to their own static interests.

Experimental Results

The framework was tested on both synthetic data and massive real-world datasets from Twitter and Reddit:

  • Twitter: Analyzing 52 million users and 1.7 billion tweets across themes like the Brazilian and Iranian elections, the study found that between 53% and 82% of users adjusted their posting behavior based on the feedback they received.
  • Reddit: In a dataset of 5 million users and 226 million comments, the numbers were lower but still significant, with 28% to 43% of users showing feedback-sensitive behavior across categories like "Leisure," "Sports," and "Learning."

These results suggest that for a vast majority of active social media users, the desire for validation and social engagement mathematically outweighs their own personal interest in specific topics.

Limitations and Future Directions

While the study provides a robust framework for understanding user behavior, it acknowledges several limitations:

  1. Static Followers: The model assumes a user's follower list remains the same, whereas in reality, followers "churn" (unfollow) if they are unhappy with the content.
  2. Linear Utility: The researchers used a linear utility function for simplicity, though human behavior may be governed by more complex, non-linear motivations.
  3. One-Way Influence: The study focuses on how users adapt to followers, but in reality, users and followers likely co-evolve, influencing each other's tastes over time.

By framing social media posting as a mathematical optimization problem, this research offers a new lens through which to view digital influence and the mechanics of online popularity.

Summary updated

Abstract

Whenever a social media user decides to share a story, she is typically pleased to receive likes, comments, shares, or, more generally, feedback from her followers. As a result, she may feel compelled to use the feedback she receives to (re-)estimate her followers' preferences and decides which stories to share next to receive more (positive) feedback. Under which conditions can she succeed? In this work, we first look into this problem from a theoretical perspective and then provide a set of practical algorithms to identify and characterize such behavior in social media. More specifically, we address the above problem from the viewpoint of sequential decision making and utility maximization. For a wide variety of utility functions, we first show that, to succeed, a user needs to actively trade off exploitation-- sharing stories which lead to more (positive) feedback--and exploration-- sharing stories to learn about her followers' preferences. However, exploration is not necessary if a user utilizes the feedback her followers provide to other users in addition to the feedback she receives. Then, we develop a utility estimation framework for observation data, which relies on statistical hypothesis testing to determine whether a user utilizes the feedback she receives from each of her followers to decide what to post next. Experiments on synthetic data illustrate our theoretical findings and show that our estimation framework is able to accurately recover users' underlying utility functions. Experiments on several real datasets gathered from Twitter and Reddit reveal that up to 82% (43%) of the Twitter (Reddit) users in our datasets do use the feedback they receive to decide what to post next.
BibTeX
@inproceedings{de2020can,
  author = {Abir De and Adish Singla and Utkarsh Upadhyay and Manuel Gomez-Rodriguez},
  title = {{Can A User Anticipate What Her Followers Want?}},
  booktitle = {WSDM},
  year = {2020},
  url = {https://openreview.net/forum?id=eLQqqLnIoJg},
}
Figure