A browser extension built for protecting users from hate speech on Twitter. Keep the web on an uplifting tone, with Uptone.
tldr; Uptone protects you from hate speech on social media by running posts through a neural network and sentiment analyzer, hiding those determined to be hateful.
Browser Extension -> Backend API -> AI Model -> Browser Extension
- Parses Tweets from page, even when more Tweets are lazy loaded on scroll
- Sends a request to our backend API to receive predictions
- Hides Tweets based on settings and model predictions, and replaces them with a custom dialog that looks native to Twitter
- Flask webserver with a POST route that serves the neural network model and sentiment analyzer
- Automatically preprocesses input and creates predictions using the model
- Sentiment analyzer complements the model by reducing false positives
- Convolutional Neural Network (CNN) that categorizes Tweet text into (0 = hate speech, 1 = offensive language, 2 = neither)
- Custom Tweet preprocessor that cleans and lemmatizes text, and expands contractions/abbreviations
- Trained on https://www.kaggle.com/datasets/mrmorj/hate-speech-and-offensive-language-dataset
- Landing page made with NuxtJS and WindiCSS, designed on Figma
cd extension
yarn install
yarn dev:chromium # or yarn dev:firefoxcd api
mkdir api/bin
cp <path to model> api/bin/hatespeech.bin
cp <path to tokenizer> api/bin/tokenizer.pickle
poetry install
FLASK_APP=api poetry run flask runFollow the instructions here to run the Jupyter notebook in notebooks/hatespeech.ipynb
cd website
yarn install
yarn dev