Back to portfolio

Deep Learning, NLP

Toxic Tweet Detection

Multi-class hate speech classifier for social media text.

Key results

86%

accuracy, SimpleRNN (best)

0.65

F1 (macro)

4

architectures compared

Deep Learning, NLP

Toxic Tweet Detection

86% accuracy, SimpleRNN (best)

Overview

A multi-class classifier distinguishing categories of toxic/hateful speech in tweets.

Topics

PyTorchKerasGloVeText Classification

Stack

  • PyTorch
  • Keras
  • SimpleRNN / LSTM / GRU / 1D CNN
  • GloVe 6B 300-d embeddings
GitHub

Key highlights

01

Compared SimpleRNN, LSTM, GRU, and 1D CNN on the same task

02

GloVe 6B 300-d pretrained embeddings

03

Best result: SimpleRNN at 86% accuracy, 0.65 macro F1

More projects

View all
Darija Word2Vec feature

Darija NLP

Darija Word2Vec

دارجةكلمشب

Competition, OpenAI Build Week

darija-gap

Speech & NLP

Meeting Intelligence System