README.md

Hate Speech Detector with Experimental CNN+LSTM

This repo is created to experiment with power of CNN on NLP tasks. The model architecture has a embedding layer(token and positional) then it is passed to CNN (1D) to get contextual meaning taking few words at one go and then again moving into next part. then these context vector is combined with another vector which is formed by context vector + embedding vector. Then these are concatenated and passed through LSTM and last layer output of hidden state is passed through sigmoid to get the classification.
This repo is experimental so model may be updated with some new addition

TODO

• Data Preparation
  • Text Preprocessing
  • Data pipeline
• Model Design
  • Encoder Head
  • Classifier Head
• Training
• Saving & Loading
  • Saving
  • Loading
• Inference