A Semi-Supervised Learning Approach to Enhance Health Care Community-Based Question Answering: A Case Study in...

Area of Science:

• Computational linguistics
• Health informatics
• Information retrieval

Background:

• Community-based question answering (CQA) sites are vital for health information but often have unanswered questions.
• Automated question answering (QA) can bridge this gap, providing valuable information to online health communities.

Purpose of the Study:

• Develop an algorithm for automatic health-related question answering using historical QA data.
• Identify key features within web-based health content that signify valid answers.

Main Methods:

• Utilized information retrieval to find candidate answers from resolved QA pairs.
• Implemented a semi-supervised learning algorithm for ranking and selecting the best answer.
• Evaluated the algorithm on a Yahoo! Answers corpus, comparing it to a string similarity baseline.

Main Results:

• The semi-supervised learning algorithm achieved 86.2% accuracy.
• Health-related features, particularly those from the Unified Medical Language System, improved performance by approximately 8%.
• Key features for identifying valid answers included text length, stop word count, question-corpus distance, and overlapping health terms.

Conclusions:

• The developed automated QA system, leveraging historical QA pairs, demonstrates effectiveness in the health domain.
• The system is designed for broad applicability across various CQA platforms.