AutoFlag (APES)

Inspiration

In collaboration with AltaML, we developed a content moderation tool for a social media platform that utilizes GPT to detect and flag hate speech, misinformation, and harmful content. One of the most un-moderated and popular social media websites on the internet to this day is 4chan, and we aim to bring our tool to help better the community moderation.

What it does

We have created a web application that flags user comments on 4chan that are offensive or harmful and marks it as such. The flagged comments are then further analyzed by a person to determine if the comment is to be deleted or ignored. This will simplify the task of moderating.

How we built it

We used Angular, python, and GPT to create a website that hosts our moderation tool.

Challenges we ran into

We initially wanted to use Twitter, but the developer API has recently changed in their usage capability, which has caused a lot of problems for previous code repositories. Therefore, given the time constraint, we decided to change to 4chan.

Accomplishments that we're proud of

We managed to get our moderation tool to be 95% accurate. Therefore the likelihood of false flags is very low.

What we learned

We learned how to build a website using Angular and GPT moderation tools to flag down inappropriate messages that are then parsed and displayed into a user accessible interface for further moderation.

What's next for AutoFlag

We plan on keeping this website up and running for the foreseeable future so that content moderation is far more accessible for existing moderators that seek a safer catalogue of content.

Fairness

Model fairness is essential to prevent discrimination and mitigate biases. GPT uses adversial training and data augmentation to maintain fairness across a wide range of inputs. The model gives a score and grades the comment for a specific category to explain the decision to flag a comment.

Social Impact

Our tool is made to create a healthier online environment for the interests of users and society.

Resourcefulness

We used OpenAI's model for their open API and their moderation criteria.

Accessibility

We decided to make a website to make our tool easy to use and able to be accessed with any device including desktop and mobile. The user interface is simple and easily understandable with only a few buttons. Also included is both a dark mode and light mode to improve quality of life.