Classifying domain-specific text documents containing ambiguous keywords.

Area of Science:

• Marine Biology
• Bioinformatics
• Computational Biology

Background:

• Keyword-based literature searches in databases like PubMed often yield irrelevant results due to ambiguous terminology.
• Manual curation by domain experts is time-consuming and essential for accurate scientific literature filtering.
• Automating this filtering process requires a solution that is fast, handles limited data, and is domain-neutral.

Purpose of the Study:

• To evaluate various classification algorithms for automating the filtering of domain-specific scientific papers.
• To develop a tool for accurately identifying echinoderm species literature from large databases.
• To assess the performance of different machine learning models in reducing irrelevant search results.

Main Methods:

• A keyword-based search was performed on comprehensive databases such as PubMed.
• Several classification algorithms were tested, including Linear, Naïve Bayes, Nearest Neighbor, Tree, SVM, Bagging, AdaBoost, and Neural Network models.
• The performance of these classifiers was evaluated for their effectiveness in filtering irrelevant articles related to echinoderm species.

Main Results:

• The developed classification tool effectively filters irrelevant articles from PubMed searches.
• The approach demonstrates high accuracy in identifying domain-specific papers, even with ambiguous common names.
• The methodology is adaptable to other fields facing similar literature filtering challenges.

Conclusions:

• Machine learning-based classification offers a practical and efficient solution for automating scientific literature filtering.
• The developed tool successfully addresses challenges posed by ambiguous keywords and limited data availability.
• This approach significantly enhances the efficiency and accuracy of literature reviews in specialized scientific domains.