VOICE-DRIVEN MEDICAL DEPARTMENT RECOMMENDATION: INTEGRATING NAMED ENTITY RECOGNITION WITH DISTINCT CLUSTERING AND MULTI-CLASS CLASSIFICATION ON UNLABELED AND LABELED DATA FOR ENHANCED ANALYSIS

Abstract

Through the integration of cutting-edge natural language processing techniques, this multidisciplinary research project seeks to improve the accuracy and reach of voice-driven medical department classification. For nuanced language comprehension, K-means clustering efficiently divides data using Clinical BERT. Integrating the Med7 and BC5CDR models helps Named Entity Recognition (NER) by making it easier to thoroughly identify and categorize things in medical records. In order to enhance contextual analysis and capture sequential relationships, LSTM networks are used for multi-class classification in conjunction with distinct BiLSTM and CRF models. To improve the classification process, a CNN model is also included to extract features from input data. This integrated framework provides useful insights into voice-driven medical record analysis and advances automated healthcare systems by working seamlessly with both labeled and unlabeled data. It has a great deal of potential to revolutionize the field. The following are the model accuracy rates: CNN 97%, CRF 78%, LSTM 76%, and BiLSTM 67%. Unlabeled patient data was grouped using K-means clustering, which shed light on how symptoms impact cluster formation and convergence.

Keywords: Medical Department Recommendation, Named Entity Recognition, K-means Clustering, Conditional Random Field, Bidirectional Encoder Representations from Transformers