Large scale biomedical texts classification: a kNN and an ESA-based approaches

Area of Science:

• Computer Science
• Bioinformatics

Background:

• Increasing textual data volume necessitates automated topic identification for document classification.
• Challenges exist in classifying documents with limited or partial text availability.
• Developing effective methods for partial information-based annotation is crucial.

Purpose of the Study:

• To propose and evaluate two novel classification methods for annotating textual documents using partial information.
• To improve upon existing k-nearest neighbours (kNN)-based approaches for this specific task.
• To assess the utility of explicit semantic analysis (ESA) as a standalone classifier and a complementary feature.

Main Methods:

• A k-nearest neighbours (kNN)-based classification approach incorporating classical Machine Learning (ML) algorithms for label ranking and additional features.
• An explicit semantic analysis (ESA)-based classification method developed as a standalone classifier.
• Investigating the combination of multiple learning algorithms and topic number determination techniques.

Main Results:

• The kNN-based method, particularly with the Random Forest learning algorithm, demonstrated strong performance on large annotated biomedical datasets.
• The kNN approach achieved a competitive f-measure of 0.55%, outperforming current state-of-the-art methods.
• The ESA-based approach yielded unsatisfactory results in this study.

Conclusions:

• Proposed classification methods are effective for annotating textual documents with partial information, suitable for large multi-label classification in the biomedical domain.
• The developed methods contribute to automated information extraction and processing of unstructured documents.
• Applications include document indexing, information retrieval, and facilitating automated data analysis.