Harmful Content Monitoring

Our project — Harmful Content Detection for Social Media Monitoring — utilises advanced data analytics and AI to automate the monitoring and flagging of harmful content, such as Hate Speech and Illicit Vape Sales, on social media platforms. Our custom AI models have also been uniquely trained to encompass Singlish, a deviation from the typical State-of-the-Art AI models.

Team members

Parwani Jayati (CSD), Yip Wen Xu (ESD), Lee Pei Xuan (CSD), Ng Tze Liang (ESD), Ernest Lim Wei Yu (CSD), Lim Hongwei Dion (CSD), Chiam Jia Hui (CSD)

Instructors:

Matthieu De Mari

Writing Instructors:

Delfinn Tan Sweimay

Why Hate Speech and Vape Sales Detection?

Hate speech on social media platforms can breed hateful and polarised communities, increase prejudice and potentially trigger instances of real-world violence. As vape sales are prohibited in Singapore, it is paramount to track and address illicit trade of vape through social media and messaging platforms in Singapore.

Our Solution

Singapore-Centric

Our models are trained to be attuned to Singlish slang and Singapore-related words.

Cross-Platform

Our dashboard supports automated monitoring across 3 social media platforms - Reddit, Youtube, and Hardwarezone Forum.

User-friendly

Informative Data Visualization with Easily Understandable Results.

How did we achieve this?

Hybrid Model Architecture

Innovated a transformer-based two-branched model architecture integrating multiple features of a social media post, ensuring robust and accurate predictions.

Each Social Media Post is evaluated holistically, i.e. our model considers multiple features of a post for predictions. The context of the post along with the comments/caption text is utilised to make a prediction. This forms the basis of our Transformer Model Architecture.

How does our solution work?

Follow 4 Simple Steps to obtain results in only a few minutes!

1. Choose Project

Select from either Hate Speech or Vape Sales detection.

2. Select Data Source

Tailor your dataset search by selecting your preferred social media platform, desired keywords, and dataset size.

3. Models In Progress

Tailored AI models designed specifically for analysing your selected dataset.

4. View and Save Results

Detailed summary of data analysis, with options to Save and Download Project.

Comparing with other models

> 90%

Ranked in Top 2

1st of its kind

Our model accuracies on both Vape Sales detection and Singapore-centric fine-grained Hate Speech classification exceed 90%.

Our F1 Scores compare favorably to the State-of-the-Art, coming in first or second on public benchmark datasets.

One-and-only Singapore Centric fine-grained Hate Speech Classification model and context-based Vape Sales detection model.

Let us create a better world by design!

Days

Hours

Minutes

Seconds

Vote the project now!

Vote for our project at the exhibition! Your support is vital in recognizing our creativity. Join us in celebrating innovation and contributing to our success. Thank you for being part of our journey!