A Knowledge-Based Semisupervised Hierarchical Online Topic Detection Framework

Area of Science:

• Computational Linguistics
• Machine Learning
• Data Mining

Background:

• Online topic models are successful but often use fixed topic numbers and suffer from increasing topic overlap.
• Hierarchical topic models offer a solution by revealing topic relationships, improving quality and reducing overlap.

Purpose of the Study:

• To propose a knowledge-based, semi-supervised, hierarchical online topic detection framework.
• To address limitations of existing online topic models, specifically fixed topic numbers and topic overlap.

Main Methods:

• Developed a novel framework for online hierarchical topic detection.
• Integrated external knowledge from knowledge sources and human experts.
• Employed semi-supervised learning and hierarchical structures.

Main Results:

• The proposed framework successfully detects topics in an online hierarchical manner.
• Experimental results demonstrate superior performance compared to baseline methods.
• Achieved competitive time efficiency alongside improved topic detection accuracy.

Conclusions:

• The knowledge-based semi-supervised hierarchical online topic detection framework effectively improves topic quality and reduces overlap.
• Integration of external knowledge enhances text mining performance in dynamic environments.
• The framework offers a promising solution for real-time topic analysis in evolving data streams.