Using Quora's Insincere Questions Dataset for Refusal Classification

When a model was built to emulate a refusal classifier, 10,000 samples were obtained and automatically classified as either a refusal or compliance./nThe dataset from the public question and answer site Quora was found to be the most suitable for the purposes of the study./nThe Quora dataset consisted of a mix of normal and offensive text strings, allowing for better identification of refusals by chat GBT./nThe Quora dataset, being in the form of a question, made it easier to determine when chat GBT was complying or not./nSome other datasets contained offensive content, but lacked a clear imperative for chat GBT to respond to.