SOTXTSTREAM: Density-based self-organizing clustering of text streams

Area of Science:

• Data Mining
• Machine Learning
• Natural Language Processing

Background:

• Density-based clustering algorithms often struggle with heterogeneous densities.
• Existing stream clustering methods frequently employ a two-phase approach (online micro-clustering, offline macro-clustering).
• The SOSTREAM algorithm introduced self-organization in the online phase to achieve single-phase macro-clustering.

Purpose of the Study:

• To present a new density-based self-organizing algorithm for text stream clustering, SOTXTSTREAM.
• To address limitations of the SOSTREAM algorithm in text data clustering.
• To demonstrate improved clustering performance using SOTXTSTREAM on real-world text streams.

Main Methods:

• Building upon the SOSTREAM algorithm's density-based, self-organizing approach.
• Implementing local density determinations to handle varying cluster densities.
• Developing SOTXTSTREAM for single-phase macro-clustering of text streams.

Main Results:

• SOTXTSTREAM effectively handles heterogeneous densities in text data streams.
• The algorithm achieves improved clustering performance compared to SOSTREAM.
• Demonstrated success on multiple real-world text stream datasets.

Conclusions:

• SOTXTSTREAM offers an effective single-phase solution for density-based text stream clustering.
• The algorithm overcomes key limitations of previous stream clustering techniques.
• SOTXTSTREAM shows significant promise for analyzing dynamic text data.