| Building on the findings from our previous work on integrating chatbot func-tionality into healthcare systems, this paper explores the intent classification task for a chatbot serving patient summaries. Our goal is to evaluate and compare the performance of various BERT-based models, including BERT, bioBERT, RoBERTa and clinicalBERT, in classifying user queries into pre-defined patient summary categories. We propose a fine-tuning approach for each model variant, using a custom dataset designed to represent the wide range of queries that patients might make about their health records. The methodology includes data pre-processing, model training, and performance evaluation across multiple metrics, such as precision, recall, and F1 score. Our experiments highlight the strengths and limitations of each model variant, providing insights into their applicability for effective patient summary query classification. The results reinforce the potential of advanced NLP models for enhancing healthcare chatbots, offering personalized and accurate interactions that improve patient experience and facilitate efficient healthcare delivery. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.