Programming techniques for improving rule readability for rule-based information extraction natural language...

Area of Science:

• Natural Language Processing
• Medical Informatics
• Software Engineering

Background:

• Extracting medical terms and values from reports is time-consuming and error-prone.
• Natural Language Processing (NLP) offers solutions for structured data transformation.

Purpose of the Study:

• Develop an NLP pipeline to extract TNM classification from pathology reports.
• Enhance the readability and maintainability of rule-based (RB) information extraction (IE) pipelines.

Main Methods:

• Manually programmed RB extraction rules for TNM classification.
• Implemented rules in two ways: direct coding and readable decomposition with intention-revealing names.
• Tested pipelines using semi-structured and unstructured pathology reports.

Main Results:

• Readable rule programming significantly reduced fine-tuning and programming time.
• Apache Uima Ruta (AURL) and Regular Expressions (REGEX) showed time savings for rule correction.
• Complex REGEX rules were reprogrammed in AURL in 5 minutes.

Conclusions:

• Improving the readability of RB IE pipelines is crucial for efficient maintenance and updates.
• Readable coding strategies facilitate understanding, transferability, and long-term maintenance of NLP pipelines.