/AggDetectApp

A web application that detects aggression and misogyny in text using BERT augmentation, sentiment analysis, XGBoost, TF-IDF vectorization, LIME explainability. [Paper accepted at ICON 2021]

Primary LanguagePythonMIT LicenseMIT

title colorFrom colorTo sdk app_file pinned python_version
AggDetectApp -- Detect Aggression and Misogyny
black
blue
docker
app.py
false
3.10.5

Agression and Misogyny Detection App

Social media platforms have become hotspots for the proliferation of trolling, aggression, and hate speech. With an overwhelming volume of social media data being generated every day, manual inspection is simply impractical. In response to this pressing issue, we present an efficient and rapid method for detecting aggression and misogyny in online social media texts.

What sets our model apart is not only its high performance but also its significantly reduced training time, model size, and resource requirements. These advantages make our model highly practical for fast inference, ensuring prompt identification of aggression and misogyny in online social media texts.

Try it out here

Features

  • Detection of Aggression and Misogyny in texts
  • LIME based prediction for explainability

Tech Stack

  • Python
  • XgBoost
  • Scikit-Learn
  • HuggingFace Transformers
  • LIME
  • Docker

Aggression Detection Results

Metric Score
F1 Score 0.735

Misogyny Detection Results

Metric Score
F1 Score 0.852

How to Run Locally

  • Clone the repo
  • Install python requirements using $ pip install -r requirements.txt
  • Run the server using $ python app.py

Additional Links