Media
5 items
Description
Scores X bots in real time. Catch suspicious profiles and replies the moment you see them, with the signals behind every call. X Bot Detector scores X (formerly Twitter) profiles and reply threads for bot likelihood in real time, helping you spot inauthentic accounts as you browse. WHAT IT DOES On any X profile, a side panel shows a bot score from 0 to 100, a HUMAN, UNCERTAIN, or BOT label, and the top features driving the score as percentages. One glance tells you how likely the account is to be a bot and which signals tipped the scale. On any reply thread, a sticky summary panel tracks how many replies are typical, possibly suspicious, and suspicious. Flagged usernames get coloured dots next to them, and flagged reply cells get coloured left borders so suspicious activity is visible at a glance. Click any count to cycle through flagged replies, or open a full report in a new tab with a flag distribution chart, the most common flagging reasons, and a sortable table of every flagged account. Click the toolbar icon at any time to see the most recent profile score or thread summary. HOW IT WORKS The profile score uses an XGBoost machine learning model trained on 37,438 labelled X accounts. The model achieves an F1 score of 0.8076 on a held-out test set. The top predictors are verified status, follower count, account age, and tweet frequency. The thread scanner uses a username-pattern heuristic because X's public reply DOM only exposes a small subset of the features the full model needs. It flags digit-heavy handles, long trailing digit sequences, and auto-generated name patterns. Verified accounts are always marked as typical. PRIVACY AND DATA HANDLING X Bot Detector does not store, log, or sell any user data. The extension reads only the profile or thread page you are actively viewing, sends a small set of public profile features over HTTPS to a Hugging Face Space for scoring, and displays the result. There is no analytics service, no tracking, and no third-party data sharing. To keep the scoring service reliable, the extension uses an anonymous access key that it obtains automatically on first use. The key is a random, machine-generated identifier used for fair-use rate limiting. It contains no personal information and is not linked to you, your browser, or your X account. The backend code is public and inspectable, and users can run the backend locally if they prefer not to use the hosted version. Full privacy policy: https://github.com/mobucheeri-datsci/x-bot-detector/blob/main/PRIVACY.md PERMISSIONS USED activeTab: read the X profile or thread the user is currently viewing in order to score it. The extension does not run on any other site. storage: cache the most recent analysis result locally so the toolbar popup can display it when reopened, and store the anonymous access key described above. Host permissions on x.com and twitter.com: the extension only operates on X. The twitter.com host is included to support users who follow old links and get redirected. Host permission on the Hugging Face Space: the backend that runs the scoring model. The extension sends extracted profile features and receives a score in return. OPEN SOURCE The model, backend, and extension are fully open source under the MIT licence. Review the code, retrain the model on different data, or self-host the backend at: https://github.com/mobucheeri-datsci/x-bot-detector LIMITATIONS The training dataset spans 2018 to 2020, so the model may not fully reflect the latest LLM-driven bot patterns. Scores should be treated as one informational signal, not a definitive verdict on any account. The model has known weaknesses on some news organisation accounts and on accounts with unusual posting patterns, the project's GitHub README documents these in detail.