This is a toxic comment classifier web application that uses a trained Logistic Regression model to predict the toxicity levels of a given text input.
Link to the web app: 👉🏻 Toxic Comment Classifier
Disclaimer: the dataset for this project contains text that may be considered profane, vulgar, or offensive.
This is a multi-label classification problem where the given input is a text comment and the output is list of the toxicity level it belongs to.
The input text data needs to be cleaned and pre-processed for it to be useful for the Machine Learning model.
The dataset for this problem was taken from competetiion hosted by Jigsaw on Kaggle.
For preprocessing of the input data and text vectorization, both word and char based TF-IDF vectorizer's output are used as inputs to the model for better performance and minimum loss of input features.
The different types of target labels present are: toxic, severe-toxic, obscene, threat, insult and identity hate.
For building the classifier, we have used Logistic Regression and treated the multi-label problem as a binary problem. The reason for this approach instead of a OneVsRest Classifier is because of better model performance when the problem is treated as a binary one.
Since the data is unbalanced, just accuary in itself cannot be considered as a strong evaluater, therefore we have used F1-score along with it to evaluate the model.
Here are the results on validation and test datasets:
Validation Results 👇🏻
Validation Accuracy: 0.9828502793879577
Validation F1-Score: 0.9811947440446507
Test Results 👇🏻
Test Accuracy: 0.9752805651942854
Test F1-Score: 0.9747181660736461
Project Structure:
Volume serial number is D8B2-80F9
D:.
├───data
│ ├───cleaned-data
│ └───raw-data
├───images
├───models
├───notebooks
├───static
│ └───css
├───templates
└───__pycache__
These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.
Flask==1.1.2
joblib==1.0.1
nltk==3.6.1
numpy==1.20.1
pandas==1.2.4
scikit_learn==1.0.2
scipy==1.6.2
swifter==1.0.9
Use miniconda to download python 3.8 or higher and then
pip install -r requirements.txt
To run the website, navigate to main folder of the project
python app.py
The server will be at "localhost:5000".
Goto "localhost:5000" and after entering the comment click on classify to predict it's toxicity values.
The model has been deployed on an EC2 instance on AWS. The IP has been made publicly accesible. Below is the link to the AWS webapp project portal:
Link: http://ec2-18-117-78-151.us-east-2.compute.amazonaws.com:8080/
Please hit the ⭐button if you like this project. 😄