Text Classification based on Sentence Features and Preprocessing Settings for Labeling eHealth Consultation Answer

This study identifies key aspects of doctor- patient communication by analyzing doctor responses from eHealth Consultation Answer data on the Online Healthcare Consultations (OHC) site, AloDokter. The analysis is performed through topic segmentation to help users better understand the doctors' explanations. Previously, researchers attempted topic segmentation on the same dataset using unsupervised learning and clustering-based text segmentation methods, but the results were suboptimal. To address this issue, we propose a solution that involves segmenting the text using text classification techniques, specifically by leveraging sentence features and refined customized preprocessing. Sentence features are enhanced by combining word vectors with sentence embeddings. The preprocessing improvements involve optimizing the tokenization process and refining stopword handling, focusing on encountered in doctor-patient communication. This research aims to compare several model text classification methods based on sentence features and preprocessing configurations and evaluate their performance. Additionally, we present the performance of text classification methods without including sentence features or customized preprocessing for comparison. Our experiments demonstrate that The Multi-Layer Perceptron (MLP) models with sentence features achieve an average F1-score of 91%, outperforming other text classification models.