The situation is the same in family circles, at work and in facebook groups. At some point someone mentions a weird fact and noone really knows what to answer. Thats why we started developing a programm that aims at identifying wrong or harmfull statements. The casual web surfer should regain trust in what she/he is reading.
What problem it solves
- Several websites publish lies as news stories. This regularly leads to the confusion of facts for open minded readers.
- Showing the people that what they are reading was proved wrong, helps immediately when people get in contact with fake-news.
What it does
- Text-Scanner: Reveal:Unreal is a database of already disproven fake news. Users provide statements, which are then fake-checked. If a synonymous fake news exists, the statement gets marked in a bright color and a link with counter proof shows up.
- Website: Currently we provide a website as a collection of all disproven statements.
- Self-improvement: Users can contribute new fake-news and rank the quality of the analysis.
How I built it
Input statements are vectorized, clustered & compared to existing entries with a similarity measure. We use current natural language processing and machine learning modules. At its core, the program learns to match semantically identical paragraphs.
The program inserts new fake-paragraphs into its database to improve accuracy and update the similarity clusters.
Challenges I ran into
Our team had to learn big parts of the techologies we use from the ground up, which defined most progress as "Ah, I get it now!" rather than "Ok, how do we want to do this?" or "technology hates me and wants to bully me!"
Accomplishments that we are proud of
- A design able to comunicte our message precisely ignoring useless functionality. It invites users to choeck their websites and statements.
- A blueprint for the machine learning process, based on a dataset of wikipedia articles and duplicate quora questions.
- A website template ready to be connected to the matching process.
- A database draft that needs to be filled with fake news data.
What we learned
All of us learned using new sofware and developement techniques. We had to adapt working as a group and keep communicating internal ideas.
What's next for Reveal:Unreal
Our team wants to actually launch the website and develop a browser extension that ships our complete AI to users.
- Clear definition of the program interfaces.
- Discussion of data protection.
- Trial & Error:
- Thrilling & user friendly frontend design.
- In depth construction of the ML process.
- More Experience:
- Developing a browser plugin for live checking of articles.