Hot Streaks on Social Media

Kiran Garimella, Robert West

Abstract: Measuring the impact and success of human performance is common in various disciplines, including art, science, and sports. Quantifying impact also plays a key role on social media, where impact is usually defined as the reach of a user’s content as captured by metrics such as the number of views, likes, retweets, or shares. In this paper, we study entire careers of Twitter users to understand properties of impact. We show that user impact tends to have certain characteristics: First, impact is clustered in time, such that the most impactful tweets of a user appear close to each other. Second, users commonly have “hot streaks” of impact, i.e., extended periods of high-impact tweets. Third, impact tends to gradually build up before, and fall off after, a user’s most impactful tweet. We attempt to explain these characteristics using various properties measured on social media, including the user’s network, content, activity, and experience, and find that changes in impact are associated with significant changes in these properties. Our findings open interesting avenues for future research on virality and influence on social media.

Introduction

Social media success is often perceived as a series of random lightning strikes. However, this research demonstrates that Twitter users experience "hot streaks"—extended periods of high engagement—driven by a sudden influx of new followers and a shift toward more diverse, media-rich content. While most social media research focuses on "single-shot" virality (why a specific tweet goes viral), this study examines the broader context of a user’s "career." Drawing inspiration from studies on "hot hands" in sports and "hot streaks" in scientific and artistic careers, the authors investigate whether impact on social media clusters in time and what factors drive these sustained periods of success.

Methodology

To analyze impact over a user's entire career, the authors employed several innovative techniques to reconstruct historical data and identify patterns.

Historical Follower Estimation

A significant challenge in longitudinal social media research is the lack of historical network data. Because the Twitter API does not provide historical follower counts, the team scraped the Internet Archive’s Wayback Machine. They used polynomial fitting (degrees 4, 5, and 6) to interpolate a user's follower count at any given timestamp in their career, as illustrated in the figure below.

Historical follower count estimation using polynomial fitting for example users.

Hot Streak Identification

The researchers treated a user's retweet counts as a time series. To identify "hot streaks," they solved an optimization problem using dynamic programming to fit a piecewise-constant function to the data. This allowed them to identify sustained periods where the "constant" level of impact was significantly higher than the user's career median.

Data Collection

The study utilized four distinct Twitter datasets to ensure generalizability:

  • VERIFIED: 2,710 high-profile accounts (celebrities, journalists, etc.).
  • POLITICAL: 10,563 users active in political discourse.
  • DESPOINA: 76,777 "standard" users filtered for minimum activity.
  • DESPOINA (RANDOM): 5,465 users sampled purely at random.

All users were required to have a "career" of at least 2,000 tweets to provide sufficient data for robust time-series analysis.

Key Insights and Results

Impact is Clustered

A user’s most successful tweets are not randomly distributed. If one examines a user's top five tweets, they are likely to occur very close to each other in time. This holds true even after normalizing for follower growth.

Pearson correlation of tweet impact over time, showing significant clustering compared to shuffled baselines.

As shown in the correlation analysis, the timing of a user's top-ranked tweets is significantly higher (roughly 0.6) than in randomized "shuffled" careers (less than 0.1). This confirms that impact is not a series of independent events but rather a phenomenon that clusters in time.

Drivers of Hot Streaks

The research identified several key characteristics of these high-impact periods:

  • The "New Blood" Phenomenon: Hot streaks are primarily driven by new retweeters. These users suddenly start engaging with a user at a high rate, though their interest is often short-lived, fading after the streak concludes.
  • Content Shifts: Users don't just tweet more during a hot streak; they tweet differently. They tend to post longer tweets, use more media (images/video), and explore a more diverse range of topics (measured by higher topic entropy).
  • The Build-up and Drop-off: There is a measurable "build-up" in engagement leading up to a user's most retweeted tweet, followed by a gradual "drop-off." Interestingly, at the very peak of impact, topic entropy actually decreases, suggesting the user focuses intensely on a specific subject right when they hit their maximum reach.

Prevalence and Predictability

Hot streaks are widespread, with 60% to 85% of users across the datasets experiencing at least one. Furthermore, these streaks are predictable; a Random Forest classifier achieved approximately 75% accuracy in identifying hot streak weeks based on features like topic entropy, followers gained, and tweet frequency.

Conclusion and Limitations

While this study provides a comprehensive look at social media careers, it is limited by its focus on high-activity users (2,000+ tweets). Additionally, while the paper identifies what changes during a hot streak, the causal direction remains an open question: does a shift in content trigger a streak, or does the momentum of a streak encourage content experimentation? Future work could extend these findings to other platforms like YouTube or Instagram, where the production barriers might influence the dynamics of sustained success.

Summary updated

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