A topic modeling framework for spatio-temporal information management

Area of Science:

• Computational social science
• Data science
• Public health informatics

Background:

• Analyzing dynamic, real-time data streams, such as social media messages, presents significant challenges due to conflicting information from diverse sources and timeframes.
• Online topic detection and tracking require robust methods to manage and process high-velocity data.
• The increasing volume of health-related discussions on platforms like Twitter necessitates effective analytical tools.

Purpose of the Study:

• To introduce a comprehensive framework for managing, processing, analyzing, detecting, and tracking topics in streaming data.
• To address the challenges of online topic detection using a novel model selector procedure with a hybrid indicator.
• To enhance data quality and improve the accuracy of health-related tweet classification.

Main Methods:

• Developed an automatic data processing pipeline with dual-level cleaning (regular and deep) incorporating meta-knowledge.
• Employed deep learning and transfer learning techniques for classifying health-related tweets.
• Integrated data visualization tools, including a US map display, for understanding topic trends over time and location.

Main Results:

• The framework demonstrated high accuracy and an improved F1-Score in classifying health-related tweets.
• The system successfully detected and tracked topics in real-time, achieving performance comparable to manual annotation.
• Graphical display on a US map effectively illustrated emerging and changing topics across different locations and time periods.

Conclusions:

• The proposed framework provides an effective solution for real-time topic detection and tracking in dynamic streaming data environments.
• The hybrid approach, combining advanced data processing with deep learning, significantly enhances the analysis of social media data for public health surveillance.
• The visualization component offers valuable insights into geographical and temporal trends of online discussions.