BotDetector автор Riley Meyerkorth

Overview

BotDetector is a open-source web extension to help users detect and flag posts that are most likely to be artificial.

Visit the repo and support me!

Features

• Per-post and per-user bot probability algorithm via a "checks and weights" system
• Checks are examinations of specific elements of a post/user. They are either true or false.
• Every check has an associated weight assigned to it. A positive weight indicates a better chance of being a bot. A negative weight indicates less of a chance to be a bot (aka a better chance to be a human/real).
• Per-post probability badges
• Extensive customization
• Customizable bot detection threshold
  • The default detection threshold is at 75%, but you can always raise or lower that to your liking!
• Per-check toggles and weight sliders in the popup
  • If you don't agree with my checks/weights, just change them!
  • In the future, I will be expanding this system to allow users to add their own checks via RegEx and other factors.
• Allowlist/blocklist
  • Editable in the popup or from a post's details dialog ("Mark as human", "Always hide").
• Show hidden posts
  • Reveals posts hidden by "Hide likely bots", and the popup shows counts for the current page.
• Export/import settings data

Detection Methods

In general, I believe in being safe than sorry. I'd rather assume the whole internet is bots than believe the amount of hate there is truly representative of reality.

Because of that belief, I believe some false-positves are worth it to help the mental health of users.

• Fancy text detection
• A lot of bots/trolls use lookalike text to confuse users
• Some users might also use unicode fonts for pazazz (i.e. "𝖈𝖑𝖎𝖈𝖐 𝖍𝖊𝖗𝖊")
• Shortend link expansion
• Links on many social media sites are shortened and/or go through a proxy. This makes it difficult for detection algorithms to find suspicious links.
• The background script safely fetches each shortened page (no cookies, no redirects followed, the destination is never contacted) and reads the destination, so adult/link-in-bio sites (OnlyFans, Fansly, bouncy.ai, beacons.ai, dm.bio...) are recognised.
• Requires host permissions for each link-shortener domain (i.e. https://t.co/*). If expansion seems to do nothing, grant it under about:addons -> BotDetector -> Permissions.
• Username history
• If the user visits the "About this user" page, a change within the last year or 3+ changes is flagged (a sign of a sold or hacked account)
• For Twitter, he extension doesn't currently fetch it itself because Twitter's private query needs rotating IDs. API support is being considered.

Keyboard Shortcuts

• Alt+R: manually re-scan reports
• Alt+K: quick-open popup menu

Developing and Contributing

To learn how to develop, build locally, and contribute to BotDetector, look at the contributing docs in CONTRIBUTING.md in the GitHub repository: https://github.com/m-riley04/BotDetector

Social Media-Agnostic Approach

Internally, this extension uses an adapter pattern approach to allow multiple different social media sites to be able to be detected.

Currently, I am focusing on Twitter (one of the main culprits for bot activity), but the system can allow for a wide variety of sites.

This system will be expanded in future versions.

AI Policy

Let me be clear: this does NOT utilize a generative AI model (i.e. ChatGPT, Claude, etc.) for it's detections.

This is NOT a ChatGPT wrapper!

Perhaps in the future, I could implement some machine learning algorithm (NOT an LLM) to detect post intent, but that's for the future.