Toxicity-Classification

This repo is about creating a simple comment toxicity classifier which is a model that takes a comment and outputs the probability of that comment being toxic (carry negative thoughts). (based on this tutorial )

The expected output is

Steps

1- Collect the dataset

2- Clean and preprocess the dataset

3- Create the model

4- Train the model

5- Evaluate the model

6- Deploy the model in the app

---

The data was downloaded from kaggle

definitely this is not a perfect classifier (there is no perfect one).

but it can get analyze comments with accuracy between 80 to 90 %

to improve these results :-

• More dataset is needed as there are a lot of neutral comments that make the model slightly biased
• Some tuning is needed to handle bias - variance trade off
• Transfer learning can be used by using BERT for a better text encoding

Prerequisites

You need to install all the following libraries to run the notebook correctly so if any library is missing, just use its corresponding command

pip install tensorflow
pip install tkinter
pip install numpy 
pip install os
pip install pandas

Testing

• Open test_app.ipynb
• Run the notebook
• Enter your sentence in the window that showed up
• Hit analyze

Built With

• Tensorflow
• Keras
• Python

Authors

• Zeyad Abdelreheem LinkedIn