Hot Streaks on Social Media

Kiran Garimella, Robert West

Abstract

While social media research often focuses on why individual posts go viral, the evolution of a creator's impact over their entire career remains less understood. This study demonstrates that social media success is not a series of independent, random events but occurs in "hot streaks"—extended periods of significantly higher-than-average impact. By analyzing decade-long Twitter careers across diverse datasets, we find that a user’s most impactful tweets are temporally clustered. These streaks are primarily driven by a sudden influx of new, short-lived followers and a temporary narrowing of topical focus.

Introduction

In fields like science and art, researchers have long observed "hot streaks," or bursts of high-quality work. This paper investigates whether similar patterns exist on social media. Is impact a "random walk" where success can strike at any time, or is it structured? Distinguishing between "one-hit wonders" and sustained periods of influence has significant implications for marketing, influencer identification, and understanding digital creativity.

Methodology

To analyze impact over time, the researchers developed a robust framework to account for platform limitations and user growth:

  1. Historical Follower Estimation: Because the Twitter API does not provide historical follower counts, the authors scraped profile snapshots from the Internet Archive Wayback Machine. They fitted polynomial curves to these data points to estimate a user's follower count at the exact time of any given tweet.
Historical follower count estimation using Wayback Machine snapshots and polynomial fitting for users @goonth and @marcno.
  1. Impact Normalization: To isolate performance from general account growth, the study used "retweets per follower" as the primary metric. This allows for a fair comparison between a tweet from a user's early career and one from their peak.
  2. Hot Streak Identification: Careers were modeled as time series of retweet counts. Using piecewise-constant function optimization, the researchers identified segments of consistently high impact while applying regularization to avoid over-segmentation.

Datasets

The study utilized four distinct Twitter datasets spanning over a decade:

  • VERIFIED: 2,710 high-profile users (50k to 2M followers).
  • POLITICAL: 10,563 users active in political discourse.
  • DESPOINA & DESPOINA (RANDOM): Approximately 82,000 "normal" users to ensure findings generalize beyond celebrities.

Key Insights and Results

Success is Clustered

A user’s top-performing tweets are not randomly distributed throughout their career. Statistical analysis reveals that if a user has a viral hit, their next major success is likely to occur shortly thereafter. The temporal correlation between the timing of a user's #1 and #2 most impactful tweets is remarkably high across all datasets.

Correlation and R² values for the temporal positioning of top-performing tweets across Verified, Political, and Despoina datasets, demonstrating strong clustering.

The Mechanics of a Streak

  • New Retweeter Engine: Hot streaks are fueled by rapid follower growth. These new followers engage at high rates initially but are often "fickle," with their engagement dropping off significantly once the streak concludes.
  • Topical Narrowing: At the peak of a hot streak, users often show decreased topic entropy. This suggests that achieving maximum impact involves focusing intensely on a specific niche rather than tweeting broadly.
  • Prevalence and Predictability: Hot streaks are near-universal for successful users, appearing in 86% of verified accounts. Using a Random Forest classifier, the researchers could predict hot streaks with approximately 75% accuracy based on network changes and content shifts.

Limitations and Future Work

The study was constrained by API limits, restricting the analysis to users with roughly 3,000 tweets. Additionally, while the study identifies a correlation between topical narrowing and success, the direction of causality remains unclear: do users narrow their focus to go viral, or does the audience's response force them into a niche? Future research could explore the "contagiousness" of hot streaks within social networks and how these patterns vary across different algorithmic platforms like YouTube or Instagram.

Summary updated

BibTeX
@inproceedings{garimella2019hot,
  author = {Kiran Garimella and Robert West},
  title = {{Hot Streaks on Social Media}},
  booktitle = {ICWSM},
  year = {2019},
}
Figure