The use of evidence-based treatment is challenging because computers that provide answers to clinical questions (CQA) are unable to comprehend complex biological language or consider various types of observations. Due to their inability to grasp the concept of link dependence, standard text-based techniques have the potential to overlook critical contextual information and provide incorrect responses when applied to biological entities. Biomedical Knowledge Graph Embedding with Transformers (Bio-KGET) is the name of the clinical quality assurance system we developed to aid in overcoming these challenges. Using electronic health records (EHRs), academic publications, and ontologies, this approach enables the construction of a biological knowledge network by integrating entities, connections, and a shared vector space. To enhance contextual reasoning and response generation, we can combine graph-structured representations with semantic embeddings in a transformer-based encoder. This improves contextual reasoning and generates more accurate responses. When applied to benchmark CQA datasets, the model outperforms baseline neural and graph-only models in terms of relevance, precision, and accuracy. Also, our model is more accurate. Based on the results, it appears that combining transformers and knowledge graph embeddings has the potential to enhance the usability, scalability, and interpretability of intelligent biomedical question-answering systems.
Alneamy et al. (Thu,) studied this question.