Mining Features Associated with Effective Tweets

Jian Xu, Nitesh V. Chawla

Abstract

This study analyzes 122 million engagements across 2.5 million tweets from 258 influential accounts to identify features driving tweet effectiveness. By normalizing engagement against follower counts, the researchers uncover non-linear relationships between tweet characteristics—such as URL count, length, and timing—and user response.

Introduction

Traditional studies of social media virality often equate success solely with retweet counts and assume linear benefits for features like tweet length or frequency. This research addresses these gaps by introducing a holistic "Effectiveness" metric that incorporates favorites and replies, normalized by follower count. The goal is to identify the specific "Goldilocks zones" where tweet composition is optimized for maximum impact.

Methodology

The authors utilized the Twitter Streaming API to collect a continuous 18-month history of high-profile accounts. They defined a core metric, Effectiveness ($E$), calculated as:

$$E(t, u) = \frac{Retweets + Favorites + Replies}{Followers}$$

This normalization ensures that engagement is measured relative to an account's reach, allowing for a fair comparison between global brands and niche influencers.

Dataset

The study monitored 258 influential accounts from November 2013 to April 2015, categorized into four groups:

  • Media Outlets: (e.g., @nytimes, @WSJ)
  • S&P 500 Companies: (e.g., @Starbucks, @Walmart)
  • CEOs: (e.g., @elonmusk)
  • Famous Investors: (e.g., @Carl_C_Icahn)

Key Insights: The Non-Linearity of Success

A primary finding is that more content does not always lead to more engagement. For instance, URL usage follows a distinct non-linear pattern.

Effectiveness peaks at 3 URLs before declining.

As shown above, effectiveness peaks when a tweet contains three URLs. Adding a fourth URL causes engagement to plummet, likely because the tweet begins to resemble spam. Similarly, the study found a "U-shaped" curve for tweet length: both very short (<20 characters) and very long (>115 characters) tweets outperform mid-length ones, as they serve distinct purposes of branding and context, respectively.

Temporal Dynamics and the Weekend Boost

Timing plays a critical role in effectiveness. The researchers identified a significant "Weekend Boost," where tweets posted on Sundays are roughly 30% more effective than those posted on Thursdays.

Sunday shows the highest average effectiveness despite lower tweet volume.

Despite this higher effectiveness, many accounts—particularly media outlets—maintain a higher volume of posts during standard 9-to-5 working hours, missing the "prime time" window (7 PM to 11 PM) when audience engagement is at its peak.

Content and Structural Drivers

  • Visual Impact: Including a picture makes a tweet 4.5 times more effective. However, the impact of GIFs and videos is polarized, helping corporate accounts but potentially detracting from the "serious" image expected of CEOs and investors.
  • Punctuation and Sentiment: Tweets with exclamation marks are 2.3 times more effective, while question marks provide a 25% boost. Furthermore, tweets expressing strong positive or negative sentiment significantly outperform neutral ones.
  • Hashtag Placement: Integrating hashtags into the middle of a sentence is 40% more effective than placing them at the end.

Limitations

The study focuses on high-profile, verified accounts, meaning the findings may not generalize to the average Twitter user. Additionally, the results indicate correlation rather than direct causality. Future research aims to incorporate more granular data, such as click-through rates, to further refine the understanding of digital influence.

Summary updated

BibTeX
@inproceedings{xu2017mining,
  author = {Jian Xu and Nitesh V. Chawla},
  title = {{Mining Features Associated with Effective Tweets}},
  booktitle = {ASONAM},
  year = {2017},
  doi = {10.1145/3110025.3110126},
}
Figure