FakeNet

• 

  Selected clusters of our graph.

• 

  A cluster of misinformation on Gaza

• 

  A part of the graph

The code : https://github.com/Jaymun723/polymagiciens/

Inspiration

One of our team member's worked during an internship in intelligence services and needed to recognise networks of influence. The idea came from this particular need.

What it does

It scraps social media to populate a database then attributes a score to each post and comment based on its trustworthiness. To compute trustworthiness we use a swarm of agents, one to evict the post not interesting, one to extract the facts discussed, one to fact-check it, etc. When those scores are computed, a proximity graph is instantiated to detect clusters of user spreading misinformation.

How we built it

The scripts for scrapping, fact-checking, computing scores and graphs run on an AWS EC2 instance. The database storing the scraped data and the scores is a PostgreSQL database hosted on AWS RDS. For the fact checking we used a Mistral Agent, for the relevance score a Nemo Agent and a PyTorch model deployed to AWS SageMaker AI. The graph visualisation tool used is Cytoscape.

Challenges we ran into

We spend a lot of time finding the right social media: our choice must be easily scrapped but a place where disinformation takes place. X (Twitter) and Facebook were our first choice but the scrapping is extremely limited. Deploying on the AWS, VPC are hard !

Accomplishments that we're proud of

Identifying a real network of disinformation about Covid vaccines on Reddit!

What we learned

The AWS SageMaker AI capabilities, multi agents infrastructures and the fact that disinformation is ubiquitous.

What's next for FakeNet

Scrapping other websites, newspaper links, analysing images and a better frontend.

Built With

• agent
• amazon-ec2
• amazon-web-services
• mistral
• networkx
• postgresql
• python
• pytorch
• sagemaker