Medical data mining in sentiment analysis based on optimized swarm search feature selection.

Area of Science:

• Computational linguistics
• Machine learning
• Medical informatics

Background:

• Sentiment prediction is vital for extracting insights from unstructured medical texts.
• Traditional text mining creates high-dimensional sparse matrices, hindering accurate sentiment prediction model induction.
• Feature selection is crucial for dimensionality reduction and improving prediction model accuracy.

Purpose of the Study:

• To propose a novel technique, optimized swarm search-based feature selection (OS-FS), for enhanced classification accuracy in medical text mining.
• To optimize swarm search using a new feature evaluation technique, clustering-by-coefficient-of-variation.
• To address challenges posed by high-dimensional sparse matrices in medical sentiment prediction.

Main Methods:

• Developed OS-FS, a swarm-type search function for ideal feature subset selection.
• Introduced clustering-by-coefficient-of-variation for optimizing the swarm search.
• Applied the method to 279 medical articles on 'meaningful use functionalities' to recognize multi-class sentiments (positive, mixed-positive, neutral, negative).

Main Results:

• Experimental results demonstrate the superiority of OS-FS over traditional feature selection methods.
• OS-FS effectively identifies optimal feature subsets, enhancing sentiment prediction accuracy.
• The method successfully handles multi-class sentiment recognition in medical text data.

Conclusions:

• OS-FS is a highly effective technique for feature selection in medical text mining.
• The proposed method significantly improves sentiment prediction accuracy compared to existing approaches.
• OS-FS offers a robust solution for analyzing complex medical literature.