Semantic Search Engine Within Anatomy Books: A BERT-Based Model for Medical Students

Medical students constantly learn Anatomical Knowledge but often lack direct access to the appropriate documentation. A semantic search tool (i.e., Question-and-Answer) will be helpful. This paper introduces a Bert-based approach designed to significantly enhance medical students’ learning experiences by swiftly guiding them to specific documents or pages containing the answers to their queries using a corpus of medical books extensively used by students as references. At the core of our semantic search engine, a fine-tuned model is proposed based on CamemBert-Bio to support question-answer interactions with medical students, particularly focusing on anatomy. The proposed model establishes a new state-of-the-art accuracy of 96.03%. We show detailed performance results and demonstrate that the proposed attention mechanism not only improves performance but also allows inspecting what regions of the question text were taken into consideration by our siamese network architecture when outputting a given vector embedding.