A Multi-View Fuzzy Clustering Framework for Semantic-Rich Text Data Using SBERT and Ensemble Learning

A Multi-View Fuzzy Clustering Framework for Semantic-Rich Text Data Using SBERT and Ensemble Learning

The increasing volume of text data across diverse fields presents substantial challenges for effective clustering and analysis. Traditional methods often struggle to capture the nuanced semantic relationships and high dimensionality of textual data, particularly in noisy or heterogeneous datasets. T...

Full description

Saved in:
Bibliographic Details
Main Authors: Mangsor Nik Siti Madihah Nik, Nasir Syerina Azlin Md, Abdul-Rahman Shuzlina, Mohemad Rosmayati
Format: Article
Language:English
Published: Sciendo 2025-01-01
Series:Applied Computer Systems
Subjects:
dimension reduction
ensemble bagging
fcm clustering
sbert embedding
Online Access:https://doi.org/10.2478/acss-2025-0011
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The increasing volume of text data across diverse fields presents substantial challenges for effective clustering and analysis. Traditional methods often struggle to capture the nuanced semantic relationships and high dimensionality of textual data, particularly in noisy or heterogeneous datasets. This study introduces a refined clustering approach leveraging a multi-view ensemble method that integrates Sentence-BERT embeddings, bootstrap bagging, and Fuzzy C-Means clustering. Multiple SBERT embeddings are initially generated to capture various facets of the text data. These embeddings are then aggregated using bootstrap bagging to enhance representation robustness. Dimensionality reduction, using Uniform Manifold Approximation and Projection (UMAP), facilitates visualization and improves cluster analysis. Finally, Fuzzy C-Means clustering is applied to identify nuanced clusters within the data. Evaluation using established metrics like the Silhouette score (0.5205), Davies-Bouldin Index (0.51), and Calinski-Harabasz Index (1 386 143.83) demonstrates significant performance improvements compared to previous methods. These findings hold potential implications for tasks such as topic modelling, sentiment analysis, and information retrieval across various text-based applications. This approach offers a promising solution for navigating the complexities of high-dimensional text data analysis.
ISSN:2255-8691

Similar Items