Skip to content

This repository contains a trained NLP classifier for hate speech and offensive language and also a chat application implemented as a Django app to demonstrate the classifier

Notifications You must be signed in to change notification settings

j95io/NLP-project-2019

Repository files navigation

About

This is a trained NLP classifier for hate speech and offensive language and a chat application that demonstrates it in real-time.

Demo

Go to chat.j95.io to try it out yourself. Type some messages into the chat and see if offensive messages, hate speech or normal messages get classified correctly.

Demo Chat Example

It's mobile friendly and you automatically get an identity assigned based on your IP. You can chat with who ever else is in the chat room at the time. Your messages will then be labeled in real time.

Features also include relabeling messages based on the larger context of what a user wrote. If this feature: "Automatic relabeling feature" is activated, the system takes the broader context of what you said over multiple messages into account.

Example: If you type "You" and in the next message "Suck", these to messages individually are not considered offensive, but taken together they are.

Implementation

Classifier

We used the ULMFiT transfer learning model, which was pre-trained on Wikipedia articles to understand language. It was then trained on labeled bullying data from Formspring and Twitter.

Demo Chat App

The chat app that we built to demonstrate the classifier with real-time labeling of chat messages is implemented as a Django app.

About

This repository contains a trained NLP classifier for hate speech and offensive language and also a chat application implemented as a Django app to demonstrate the classifier

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published