RespectNet

RespectNet is a new way to bring kindness into the age of cyberbullying. Our aim is to not only rephrase rude, mean, or hurtful comments that school aged children could post to one another, but also teach them the impact their words could have and why they should change them. We want to bring forth a solution that will improve how school aged children speak with one another online and hope the learnings they are taught can be transferable to their daily life as well. This will be done through a user interface in which the user can enter the comment they plan to post to social media or in a text message, then they will get an output that returns a rephrased comment if required and an explanation of why their original comment needed to be rephrased. We’ve accomplished this by leveraging three key LLMs - Jigsaw’s Perspective, OpenAI’s Moderation, and Cohere. This has been broken down into two modeling types - classification and generative. Through the classification model we are able to utilize Jigsaw’s Perspective and OpenAI’s Moderation to intake the comments from UC Berkeley’s DLab: Measuring Hate Speech database, filtering and categorizing these comments using OpenAI’s Moderation, then evaluating the toxicity of these comments with Jigsaw’s Perspective. In the generative model we will be intaking the user’s comment and creating the corresponding output using Cohere and a prompt we’ve engineered for the RespectNet goals