It Doesn't Break Just on Twitter. Characterizing Facebook content During Real World Events

Prateek Dewan, Ponnurangam KumaraguruarXiv 2014 · arXiv

Introduction

For years, the computer science research community has operated under a "Twitter-centric" paradigm. Twitter's public-by-default nature made it the primary laboratory for studying information diffusion during crises, elections, and major cultural moments. Facebook, despite its significantly larger user base, was often overlooked due to its "private-by-default" architecture, which complicates large-scale data collection. This paper challenges the assumption that Facebook is a secondary player in real-time news, demonstrating that its public content is a rich, timely, and largely untapped resource for event tracking.

Methodology

The researchers employed the MultiOSN framework to monitor 16 major real-world events over a nine-month period in 2013. These events spanned various categories, including natural disasters (Oklahoma Tornado), sporting events (IPL), and breaking news (Boston Marathon Blasts). To ensure a rigorous comparison, identical keyword-based searches were used to collect data from both Facebook and Twitter.

The technical analysis focused on three primary dimensions:

  1. Textual Overlap Analysis: Using the Porter Stemmer to normalize text, the authors compared the core conversational themes across platforms to find shared discourse.
  2. Spam Identification: The team developed a classification system using Random Forest, J48, and Naive Bayes algorithms. They introduced novel stylometric features, such as post and hashtag repetition factors, to distinguish automated spam from organic user content.
  3. Identity Linking: By tracking identical URLs posted simultaneously across platforms, the researchers sought to identify cross-platform user behavior.

Comparative Analysis: Volume and Timeliness

A common misconception is that Facebook lags significantly behind Twitter in reporting breaking news. However, this study found that for unprecedented events, the first relevant Facebook post appeared within an average of 11 minutes. This drastically contradicts previous estimates suggesting a 10-hour latency.

Figure 1: Comparison of public content volume on Facebook and Twitter across 16 major events in 2013.

While Twitter often produced a higher raw volume of tweets per event, the sheer scale of Facebook means that even its small percentage of public posts (~28%) results in approximately 1.33 billion posts daily—dwarfing Twitter’s total output of ~500 million. As seen in Figure 1, the activity levels on both platforms often mirror each other, though specific events like the Navy Yard Shooting saw Facebook activity spikes that rivaled or exceeded Twitter's relative trends.

The "Twitterization" of Facebook and Cross-Platform Behavior

The study highlights a fascinating trend of "Twitterization" on Facebook. Users were observed using hashtags on Facebook well before the platform officially supported them in June 2013. This behavior was largely driven by cross-posting via third-party applications.

Figure 2: Distribution of URL sources on Facebook (a) and Twitter (b), highlighting the prevalence of cross-platform tools like HootSuite.

Analysis of URL sources (Figure 2) reveals that tools like HootSuite are major contributors to content on both platforms. The researchers found that at least 70% of identical URLs posted at the same time on both networks originated from the same user, providing a reliable method for linking digital identities across the social web. Furthermore, approximately 30.56% of unique textual content on Facebook overlapped directly with Twitter during major events.

Spam and Content Quality

The nature of spam differs significantly between the two platforms. Twitter spam is frequently characterized by "topic hijacking," where trending hashtags are used to promote products. In contrast, Facebook spam tends to be more "radical and polar," often involving religious or political propaganda. Despite these differences, the researchers' stylometric detection model achieved remarkable success, with the Random Forest classifier reaching 99.9% accuracy on Facebook and 98.2% on Twitter.

Conclusion and Future Work

This research proves that Facebook is not just a social network for private interactions but a vital component of the global real-time information ecosystem. By ignoring Facebook, researchers miss nearly a third of the unique public discourse surrounding major events.

Future research should aim to categorize events more granularly to determine if platform dynamics shift based on the nature of the crisis (e.g., natural disasters vs. political upheavals). Additionally, overcoming the "Privacy Wall" remains a challenge; exploring methods to extract network-level propagation data within the constraints of modern APIs will be crucial for a more complete understanding of how information moves through the world's largest social network.

Summary updated

BibTeX
@misc{dewan2014it,
  author = {Prateek Dewan and Ponnurangam Kumaraguru},
  title = {{It Doesn't Break Just on Twitter. Characterizing Facebook content During Real World Events}},
  year = {2014},
  eprint = {1405.4820},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1405.4820},
}
Figure