Related Experiment Videos
Biomedical literature classification using encyclopedic knowledge: a Wikipedia-based bag-of-concepts approach
Marcos Antonio Mouriño García1, Roberto Pérez Rodríguez1, Luis E Anido Rifón1
1Department of Telematics Engineering, University of Vigo , Vigo , Spain.
Peerj
|October 16, 2015
Summary
This study introduces a new method for classifying biomedical documents using concepts from Wikipedia, outperforming traditional word-based methods. This approach significantly improves information retrieval for researchers by overcoming issues like synonymy and polysemy.
Area of Science:
- Natural Language Processing
- Bioinformatics
- Information Retrieval
Background:
- Automatic classification of biomedical literature is crucial for researchers managing vast amounts of information.
- Traditional bag-of-words (BoW) models struggle with synonymy and polysemy, limiting classification accuracy.
- Existing methods rely on word frequency, not semantic meaning, for document representation.
Purpose of the Study:
- To evaluate a novel classifier that utilizes Wikipedia's encyclopedic knowledge for biomedical document classification.
- To introduce a bag-of-concepts (BoC) representation that addresses synonymy and polysemy by focusing on semantic units.
- To compare the performance of BoC against traditional BoW representations in biomedical text classification.
Main Methods:
- Developed a bag-of-concepts (BoC) document representation using Wikipedia as an external knowledge source.
- Implemented a classification strategy that weights concepts based on their semantic relevance within the text.
- Conducted empirical evaluations on the OHSUMED and UVigoMED biomedical corpora.
Main Results:
- The Wikipedia-based BoC representation significantly outperformed the classical BoW representation across both corpora.
- Achieved up to 157% improvement in single-label and 100% in multi-label classification on the OHSUMED corpus.
- Demonstrated up to 122% improvement in single-label and 155% in multi-label classification on the UVigoMED corpus.
Conclusions:
- Leveraging Wikipedia for a bag-of-concepts representation enhances biomedical document classification accuracy.
- The BoC approach effectively tackles synonymy and polysemy, offering a more robust method than BoW.
- This semantic-based classification strategy provides a more effective way for biomedical professionals to access relevant literature.