RingBDStack/SocialED

What it solves

SocialED addresses the difficulty of deploying and evaluating social event detection (SED) models due to the lack of a unified framework. It provides a standardized Python toolkit that integrates numerous algorithms and datasets, simplifying the process of identifying and analyzing real-world events (such as natural disasters or viral phenomena) from social media platforms like Twitter and Facebook.

How it works

Built on top of PyTorch and DGL, the library implements a unified API consisting of four primary stages:

1. Preprocess: Handles graph construction and tokenization.
2. Fit: Trains the detection algorithms and adjusts model parameters.
3. Detection: Identifies events from the input data using the trained model.
4. Evaluate: Measures performance using metrics like precision, recall, and F1-score.

Who it’s for

It is designed for AI researchers and practitioners in the field of Natural Language Processing (NLP) and social media analysis who need a consistent platform to benchmark different SED algorithms or deploy them for applications like crisis management, public opinion monitoring, and fake news detection.

Highlights

• Extensive Algorithm Library: Includes 19 detection algorithms ranging from classic methods (LDA, Word2Vec) to advanced GNNs (KPGNN, QSGNN) and PLMs (BERT, SBERT).
• Diverse Dataset Support: Integrates 15 datasets covering multiple languages (English, French, Arabic) and scenarios (general events, catastrophic crises).
• Flexible Architecture: Offers modular components and a consistent interface for seamless integration across different models.
• Hardware Support: Supports both CPU and GPU for high-performance training and inference.