Semantic Relatedness Graph for Text Segmentation of Patient-Centered Communications in Question-Answer Data

In this research, we explore text segmentation in patient-centered communications within online health consultations using the Semantic Relatedness Graph (SRG) method. The study is motivated by the need to improve the clarity and understanding of doctor-patient interactions in online communication environments, where patients often struggle to understand medical advice. The SRG method segments doctor responses by grouping semantic related sentences, making it easier for patients to follow the explanations given. We used an Indonesian health consultation dataset from Alodokter.com and categorized communications into six key patient-centered communication functions, as defined by Ann King. This six-function model is crucial as it covers various aspects of medical interactions, such as relationship building, information sharing, decision-making, and emotional support, ensuring that the segmentation approach captures the important elements of healthcare communication. We compared the SRG method to Latent Dirichlet Allocation (LDA) for topic-based segmentation, evaluating their performance using Cohen's Kappa to measure agreement between automated segmentation and manually annotated data. While the SRG method identifies related sentences, it shows some limitations in matching the predefined medical communication categories, unlike LDA, which has slightly better alignment. However, the SRG method's ability to detect semantic related sentences indicates its potential for further refinement in healthcare communication analysis.