Skip to content

Welcome to DSA4264 Team Detoxify Documentation

Members: Joy Tan, Kylie Tan, Koh Yi Jing, Richmond Sin, Sarah Goh

Welcome to the documentation for Team Detoxify in DSA4264, where we focus on analyzing and mitigating toxic and hateful online comments. Our project leverages advanced machine learning models to classify harmful content and uncover underlying themes, enabling safer and more constructive digital interactions.

Report Overview

Our technical report is structured as follows:

1. Context

  • Overview of the problem: the prevalence and impact of toxic and hateful online content.
  • Motivation for the project and its significance in creating safer digital environments.

2. Scope

  • Defines the boundaries of our project, including key objectives and deliverables.
  • Highlights the specific focus areas within toxicity and hate speech classification.

3. Methodology

3.1 Data Processing

  • Data collection and cleaning processes.
  • Description of the datasets used, including their sources, size, and preprocessing steps.

3.2 Modelling/Multiclass Text Classification

  • Development and fine-tuning of our Multilingual DistilBERT model.
  • Explanation of the classification categories and metrics used to evaluate model performance.

3.3 Modelling/Topic Modelling

  • Application of BERTopic to identify and analyze key themes in toxic comments.
  • Explanation of how BERTopic extracts topics and provides deeper insights into the nature of harmful content.

3.4 Application

  • Implementation of the model in a real-world setting through our Streamlit application.
  • User workflows for classifying and analyzing comments, including manual input and bulk analysis via CSV uploads.

4. Findings

  • Key insights and trends derived from the analysis.
  • Visualizations and interpretations of the results, highlighting the effectiveness of our models in detecting and categorizing toxic content.

This documentation provides a comprehensive guide to our methodologies, findings, and applications, serving as a reference for understanding and extending our work.