Dynamic Sub-Swarm Approach of PSO Algorithms for Text Document Clustering.

Area of Science:

• Data Mining
• Artificial Intelligence
• Computational Intelligence

Background:

• Text document clustering is vital for applications like IoT data retrieval and duplicate detection.
• Swarm intelligence (SI) algorithms, particularly particle swarm optimization (PSO), show promise for complex text clustering.
• Standard PSO can suffer from premature convergence to local optima, limiting its effectiveness.

Purpose of the Study:

• To propose a novel dynamic sub-swarm PSO (subswarm-PSO) algorithm for text document clustering.
• To enhance the global search capabilities of PSO and mitigate the issue of local optima.
• To evaluate the performance of subswarm-PSO against standard PSO and K-means algorithms.

Main Methods:

• Implementation of a dynamic sub-swarm strategy within the PSO framework.
• Comparative analysis using six benchmark datasets.
• Performance evaluation using the purity metric and execution time.

Main Results:

• The proposed subswarm-PSO algorithm achieved higher purity scores than standard PSO and K-means.
• Subswarm-PSO demonstrated improved global search capabilities, effectively avoiding local optima.
• The execution time of subswarm-PSO was found to be slightly less than that of standard PSO.

Conclusions:

• The dynamic subswarm-PSO approach offers a significant improvement for text document clustering.
• This method provides a more robust and efficient solution compared to traditional algorithms.
• Subswarm-PSO effectively addresses the limitations of standard PSO in complex clustering tasks.