SafeSpeak

Inspiration

We identified a variety of issues that inspired us to build this project: social media is rampant with hate speech, and human moderators are often overworked due to the psychological effects of reading hate speech and flagging it for their jobs, causing massive turnover and burnout. Further, there are types of hate speech that are not overt, utilizing dogwhistles through acronyms, phrases, and numbers to connect with others who share similar ideologies. To tackle the amount of hate speech online, we created an AI-powered hate speech identifier that companies can utilize to take the pressure off of their moderators and create a safer and more inclusive platform. Further, our program can help promote safety within online communities by preventing the spread of harmful and even dangerous ideas.

What it does

Utilizing AI, our program scans text and identifies instances of hate speech and intricate dogwhistles associated with hate groups.

How we built it

We utilized OpenAI's API, Python for the backend, Python Pandas for data collection and prompt engineering, and Flask for the front end design.

Challenges we ran into

Collecting data, as well as training the AI to properly detect words/phrases for flagging was the most difficult part.

Accomplishments that we're proud of

We were able to use web scraping to compile our own data from a variety of online sources and integrate it with a large language model to accurately identify hate speech within text, including scanning the text for hateful undertones or context rather than just simply identifying if the word exists within the text.