Related Experiment Videos
Utilizing weakly controlled vocabulary for sentence segmentation in biomedical literature
1School of Knowledge Science, Japan Advanced Institute of Science and Technology. ken@jaist.ac.jp
In Silico Biology
|June 24, 2005
Summary
This study introduces an improved method for building biomedical dictionaries and matching terms, addressing challenges like null boundaries and word variations. The approach successfully identifies a significant portion of meaningful terms in biomedical texts.
Area of Science:
- Biomedical Informatics
- Natural Language Processing
- Computational Linguistics
Background:
- Biomedical texts contain numerous domain-specific terms, necessitating effective term matching.
- Challenges in biomedical term matching include null boundaries between terms and the infinite variations of generative words.
Purpose of the Study:
- To develop an efficient approach for building comprehensive biomedical dictionaries.
- To implement a robust term matching system capable of handling complex linguistic variations in biomedical texts.
Main Methods:
- Gathered a large corpus of biomedical terms, including those with part-of-speech (POS) and category information.
- Utilized a completion program to generate approximately 1.36 million term variants, mitigating stemming issues.
- Stored the dictionary in a relational database management system (RDBMS) for rapid lookups.
- Developed a matching program that overcomes null boundary limitations and accommodates generative words.
Main Results:
- Experimental results on the GENIA corpus demonstrated promising performance.
- Nearly half of the potential terms were accurately recognized as meaningful segments.
- A significant portion of the remaining terms could be identified through post-processing techniques like chunking.
Conclusions:
- The developed dictionary building and term matching approach effectively addresses key challenges in biomedical text processing.
- The system achieves reasonable segmentation and dictionary lookup performance without relying on term cost, connectivity cost, or syntactic information.