Using Semantic Text Similarity calculation for question matching in a rheumatoid arthritis question-answering system

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Medical Informatics

Background:

• Intelligent question-answering systems (like FAQs) provide verified answers but lack specificity for chronic diseases.
• Existing technologies for chronic disease question-answering systems are not sufficiently mature.
• Rheumatoid arthritis is an example of a chronic disease lacking dedicated question-answering systems.

Purpose of the Study:

• To develop an improved question-answering system for chronic diseases.
• To enhance the understanding of user query intent in the context of chronic disease information seeking.
• To leverage deep learning for more accurate and relevant responses.

Main Methods:

• Utilized Bidirectional Encoder Representations from Transformers (BERT) language model to embed question classification information into sentence vectors.
• Employed edit distance for initial candidate question retrieval.
• Integrated multi-head attention and fully connected feedforward layers to extract and fuse sentence features for semantic similarity calculation.

Main Results:

• Achieved Top-1 precision of 0.551, Top-3 precision of 0.767, and Top-5 precision of 0.813 on 176 test sentences.
• Demonstrated strong performance in accurately understanding user query intentions.
• The model effectively processed semantic similarities for improved question matching.

Conclusions:

• The developed deep learning model integrates question background and classification for enhanced understanding.
• Combines the efficiency of deep learning with semantic comprehensibility.
• Improves the ability of intelligent question-answering systems to grasp user intent and deliver highly relevant answers for chronic disease queries.