Text clustering based on pre-trained models and autoencoders.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Text clustering is crucial for medical decision-making, patient record management, and information retrieval.
• Traditional methods like bag-of-words struggle with high dimensionality, sparsity, and context in large medical datasets.
• Deep learning offers advanced solutions for complex, nonlinear data, outperforming traditional clustering models.

Purpose of the Study:

• To develop an advanced text clustering model for medical data.
• To address the limitations of traditional clustering algorithms in handling complex medical text.
• To improve the accuracy and efficiency of medical data analysis and retrieval.

Main Methods:

• The study integrates pre-trained language models (PLMs) with deep embedding clustering (DEC) models.
• PLMs capture sequential text information, including word positions and context.
• Autoencoders within DEC learn data representations and clustering information, reducing noise.

Main Results:

• The proposed model demonstrated superior performance on four public datasets compared to existing text clustering algorithms.
• The integration of PLMs and DEC effectively handles high-dimensional and complex medical text data.
• The model shows significant potential for practical application in medical data clustering.

Conclusions:

• Combining pre-trained language models with deep embedding clustering offers a powerful approach for medical text analysis.
• This method overcomes the limitations of traditional text clustering techniques in the healthcare domain.
• The developed model provides a robust solution for enhancing medical data management and decision support systems.