An Intelligent Drug QA System Integrating Dynamic Knowledge Graph and Enhanced Entity Recognition via Optimized Aho-Corasick and Multi-Scale TextCNN

This study proposes an intelligent drug QA system based on a dynamic knowledge graph, addressing the increasing demand for accessible healthcare information. Multi-source drug data were integrated into a temporal knowledge graph using Neo4j, enabling real-time updates such as additions, deletions, and modifications. An improved Aho-Corasick algorithm with a hash-based storage structure was implemented for drug entity recognition, achieving an average response time of under 10ms. For intent recognition, an enhanced TextCNN model, incorporating self-attention mechanisms, was employed, reaching over 90% accuracy on the test set. The entire QA process was modularly implemented in Python, achieving an overall system query accuracy of 95% and encompassing 79,785 entities and 111,692 relationships. Experimental results confirm that the proposed system efficiently and accurately retrieves drug information in real time, offering a scalable and temporally aware solution for intelligent pharmaceutical consultations.