Design, implementation, and operation of a rapid, robust named entity recognition web service

Area of Science:

• Biomedical text mining
• Computational biology
• Bioinformatics

Background:

• Traditional BioCreative tasks prioritize annotation quality (precision, recall).
• Practical text mining applications require consideration of interoperability, speed, and stability.
• Established named entity recognition (NER) web services have been operational for a decade, designed for efficiency and robustness.

Purpose of the Study:

• To adapt an existing NER web service for the BioCreative challenge.
• To evaluate the performance of the enhanced NER service, focusing on speed and stability.
• To identify potential bottlenecks in the text mining workflow.

Main Methods:

• Extended an existing NER tagging service with BeCalm API support.
• Utilized a multi-threaded queueing system for handling simultaneous requests.
• Hosted the service at a supercomputer facility for optimal performance.

Main Results:

• The enhanced NER tagger maintained high efficiency, processing 5000 abstracts in under 30 seconds.
• The service experienced no downtime during the challenge.
• Document text retrieval from challenge servers was identified as the main performance bottleneck.

Conclusions:

• The developed NER service is efficient and stable for large-scale text mining tasks.
• Document retrieval speed significantly impacts overall performance.
• Future API standards should consider local document storage or caching for improved efficiency.