FakeDetector: A measurement-based tool to get rid out of fake content in your BitTorrent Downloads

Michal Kryczka, Rubén Cuevas Rumín, Roberto Gonzalez, Ángel Cuevas, Arturo AzcorraarXiv 2011 · arXiv

TL;DR

This paper tackles the pervasive issue of fake content distribution in BitTorrent, revealing that a significant portion of shared files are malicious or lead to scams. They introduce TorrentGuard, a novel system that leverages IP address blacklisting for early detection, aiming to prevent millions of harmful downloads annually.

What Problem Does This Paper Address?

The BitTorrent ecosystem, a cornerstone of peer-to-peer file sharing, is plagued by "fake content." This isn't just about incorrect file names; it refers to torrents that, when downloaded, lead to malware infections, phishing attempts, or redirect users to scam websites. While BitTorrent portals like The Pirate Bay attempt to combat this by removing fake torrents and banning user accounts, their methods are reactive and limited to their specific platform. The authors highlight that these existing solutions are inefficient, require significant human intervention, and don't address the problem system-wide. This leaves millions of users vulnerable to computer infections and financial scams.

Methods

The core of the paper is the introduction of TorrentGuard, a system designed for early detection of fake content. Unlike existing methods that rely on usernames, TorrentGuard identifies fake publishers by their IP addresses. The system operates through several modules:

  • RSS Parser: Monitors BitTorrent portals (specifically The Pirate Bay) for new torrent announcements, extracting infohashes and publisher usernames.
  • Tracker Connector: Connects to the tracker associated with a torrent to obtain the IP address of the initial seeder. If multiple seeders are present, it analyzes bitfields to identify the original seeder.
  • PirateBay Parser: Periodically checks publisher webpages on The Pirate Bay. If an account is removed, it flags the publisher as potentially fake. An IP address is blacklisted as definitively fake after multiple associated accounts are removed (configurable, set to 3 in their study).
  • Database: Stores all collected information, including blacklists of fake torrent infohashes and fake publisher IP addresses.

TorrentGuard's key innovation is its IP-based detection, allowing it to identify fake content as soon as it's published, rather than relying on user reports after the fact. Furthermore, by using the torrent's infohash, it can identify fake content across the entire BitTorrent ecosystem, not just on a single portal.

Datasets

The study utilizes data collected from The Pirate Bay portal over a 14-day period (April 30, 2011, to May 13, 2011), including a 5-day warm-up phase. They collected approximately 29,330 torrents, of which 10,206 (35%) were identified as fake. They also collected IP addresses of peers participating in swarms associated with fake content.

Key Insights and Ideas

  • Concentrated Malice: A surprisingly small number of users (just 20 identified "Top Fake Publishers") are responsible for a vast majority (90%) of the fake content.
  • Hosting Provider Hubs: A significant portion of these fake publishers operate from Hosting Providers, suggesting they rent dedicated servers for their operations, likely for resources and anonymity.
  • Sophisticated Tactics: Fake publishers aren't just uploading random files. They employ strategies like using popular content titles, mimicking trusted users, and manipulating statistics (seeders/leechers) to attract victims.
  • Malware and Scams Dominate: Over 99% of the analyzed fake content is linked to either malware distribution or scam websites, highlighting the direct threat to users.
  • IP Address is the Key: The paper strongly argues that IP addresses, not usernames, are the true identifiers of fake publishers, as usernames can be easily created and discarded.
  • Countermeasures are Limited: Existing anonymity techniques like TOR significantly degrade performance, making them impractical for the high-volume activity of fake publishers.

Key Results

  • Prevalence of Fake Content: 35% of analyzed torrents on The Pirate Bay were fake, a 5% increase from a previous study a year prior.
  • Publisher Identification: Only 71 IP addresses were responsible for a significant portion of fake content. The top 20 IP addresses accounted for 90% of fake content.
  • Detection Time Improvement: TorrentGuard detected fake content significantly faster than The Pirate Bay. It identified around 50% of fake content at its birth, reducing detection time by a median of 60 minutes, and over 2 hours for 20% of cases.
  • Download Prevention: If widely adopted, TorrentGuard could prevent an estimated 35 million fake content downloads per year, translating to hundreds of thousands of avoided computer infections and scam incidents.

Limitations and Future Work

  • Dependency on The Pirate Bay: TorrentGuard currently relies heavily on The Pirate Bay for identifying new fake publishers. The authors propose extending TorrentGuard to support other portals with similar announcement (RSS) and reporting (account removal) mechanisms.
  • Dependency on User Reports: While TorrentGuard automates much of the process, it still relies on user reports (account removals) to definitively mark an IP as fake. They aim to minimize this dependency by using a configurable threshold for marking IPs.
  • Future Deployment: The authors envision a more integrated system where BitTorrent portals and trackers collaborate. Trackers could identify fake publisher IPs, and portals could blacklist them, creating a more robust, ecosystem-wide defense. They are actively seeking partnerships with major BitTorrent portals and trackers.

Summary updated

BibTeX
@misc{kryczka2011fakedetector,
  author = {Michal Kryczka and Rubén Cuevas Rumín and Roberto Gonzalez and Ángel Cuevas and Arturo Azcorra},
  title = {{FakeDetector: A measurement-based tool to get rid out of fake content in your BitTorrent Downloads}},
  year = {2011},
  eprint = {1105.3671},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1105.3671},
}

Loading related content...

Figure