Named Entity Recognition-based Hospital Recommendation
S Rajathi, Reddi Tharun Kumar, Sripathi Vamsi Krishna, S Sabareesh, Pandi Eswar Chand Ethihas · 2023
Extracting the intended information from raw text input remains a potential challenge in the biomedical industry. An NER based pre-diagnosis model that uses a general medical description from the patient is proposed to extract the symptoms. In addition, a Machine Learning model that predicts the disease and recommends a hospital is discussed. To develop this pipeline, we collected the BC5CDR dataset, supplemented with manually annotated data, and trained it using several models, including the spaCy blank model, pre-trained BC5CDR model, and transformer models such as the BERT case and RoBERTa. After evaluating the models' performance, we found that RoBERTa outperforms other models with a higher F1 Score of 0.8586 and precisely identifies symptoms, diseases, and chemicals from the medical description. These physical symptoms are used by ML techniques to predict possible disease. Then, using the hospital's data and predicted results, the K-nearest neighbor algorithm suggests nearby, highly rated hospitals based on the user's location, distance, and ratings. This study can potentially enhance medical diagnoses and treatments and provide a promising foundation for future medical consultations and self-diagnosis.