Local AI misinformation classification system.
My bachelor's thesis at the University of Lübeck.
My bachelor's thesis at the University of Lübeck.
Development of a browser extension for
to display warnings about misinformation
Development of a browser extension for X to display warnings about misinformation
Bachelor Thesis
My bachelor's thesis at the University of Lübeck.
Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet.
Highlights
Misinformation Classification
Striking design.
A brilliant experience.
Five verdicts, not two. Real claims rarely split cleanly into true and false. Every post is sorted into one of five categories — True, False, Disputed, Unclear or Parody — so nuance is never forced into a binary.
Measured, not just labelled. When a post is flagged false, it is scored from 0 to 10 for both truthfulness and harm potential — turning a single verdict into something you can actually weigh.
Classification Scale. One spectrum runs from False through Disputed, Unclear and Parody to True — each step carrying its own colour and icon for instant orientation.
Tap to go deeper. Progressive disclosure keeps the feed calm: a compact badge first, a one-line reason on tap, and a full detail card with analysis, red flags and sources only when you want it.
It reads the picture, too. Beyond the text, the model can describe a post's image and weigh it against the words — catching the cases where a visual quietly contradicts the claim.
Every call, explained. The detail card breaks each decision into a content analysis, the linguistic and structural red flags it noticed, and plain-text sources you can look up yourself.
Built for every kind of sight. High-contrast, colour-blind and icon-only modes keep every warning legible — and never lean on colour alone to make their point.
Some calls need no AI. Clearly declared ads and parody accounts are caught deterministically, before the model ever runs — saving compute and avoiding needless verdicts.
Evaluation
Two ways to verify.
One question of trust.
The machine explains. You decide. Every verdict can be checked from two sides — the model's own reasoning, and your personal judgment. Neither overrides the other; together they answer one question of trust.
Your feed, in numbers. A local dashboard charts how your timeline breaks down across categories and how that mix shifts over time — across every post the tool has ever analysed.
A complete archive. Every classified post is kept locally and stays browsable, so you can revisit any decision long after it scrolled out of sight.
See exactly how it decided. A dedicated reasoning view walks through the logic end to end — from prompt to verdict — answering the question every label quietly raises: how was this determined?
Nothing hidden. The prompt log shows the exact instruction sent to the model and its raw, unedited answer — full transparency over how each result came to be.
Your verdict counts too. Disagree with a label? Record your own assessment in the feedback log. It is stored only for you and never alters the model's output.
Credibility, earned over time. From at least three classified posts, an author's track record is aggregated into a credibility signal — context that no single tweet could ever give.