TextNetTopics: Text Classification Based Word Grouping as Topics and Topics' Scoring

Area of Science:

• Computer Science
• Bioinformatics
• Medical Informatics

Background:

• Medical document classification faces challenges due to large, noisy feature sets.
• Feature selection is crucial for improving classification model accuracy.
• Traditional Bag-of-Words (BOW) models can be suboptimal for complex medical data.

Purpose of the Study:

• To propose a novel feature selection approach, TextNetTopics, for medical document classification.
• To enhance classification performance by selecting relevant topics rather than individual words.
• To evaluate TextNetTopics on the CAMDA challenge dataset and other textual datasets.

Main Methods:

• TextNetTopics applies a Bag-of-Topics (BOT) approach for feature selection.
• The method is based on the G-S-M (Grouping, Scoring, and Modeling) framework.
• Topics are scored to identify the most discriminative ones for classifier training.

Main Results:

• TextNetTopics demonstrated superior performance compared to various existing feature selection methods.
• The approach achieved high performance on the CAMDA challenge validation data.
• The algorithm was successfully applied to diverse textual datasets, showing generalizability.

Conclusions:

• TextNetTopics offers an effective alternative to traditional Bag-of-Words feature selection in medical text classification.
• Topic-based feature selection can significantly improve the accuracy and robustness of classification models.
• The G-S-M framework, adapted as TextNetTopics, shows promise for biological and medical data analysis.