Improving imbalanced scientific text classification using sampling strategies and dictionaries.

Area of Science:

• Information Science
• Computer Science
• Biomedical Informatics

Background:

• Class imbalance is a common problem in scientific document classification, affecting retrieval accuracy.
• Oversampling and Subsampling are common techniques to address class imbalance.
• Effective classification of biomedical texts is crucial for scientific documentation recovery.

Purpose of the Study:

• To evaluate the impact of Oversampling and Subsampling strategies on classifiers for imbalanced scientific document data.
• To assess the effectiveness of different dictionaries in biomedical text classification.
• To optimize search performance on the PubMed scientific database.

Main Methods:

• Applied Oversampling and Subsampling techniques to address class imbalance.
• Tested three classifiers: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Naive-Bayes.
• Utilized three dictionaries: BioCreative, NLPBA, and a subset of UniProt (Protein) for text classification.
• Compared results against established benchmarks using the TREC Genomics 2005 corpus.

Main Results:

• The Subsampling technique combined with the SVM classifier achieved the best performance.
• The NLPBA and Protein dictionaries demonstrated superior effectiveness in classification tasks.
• The chosen methods showed competitive results compared to existing literature.

Conclusions:

• Subsampling is an effective strategy for handling class imbalance in scientific document search.
• Dictionary choice significantly impacts the performance of biomedical text classifiers.
• SVM classifier with appropriate dictionaries and sampling techniques offers a robust solution for imbalanced scientific data.