Federated Deep Learning for Emotion Recognition in Clinical Narratives

View at publisher Summarise with VIRAT
Authors: Ramesh S. Wadawadagi, Tej Hiremath, Shrikant Tiwari
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Emotion Recognition in clinical narratives aims to identify the emotional states expressed by patients regarding their health conditions, diagnoses, drug responses, side effects, etc., from conversation transcripts, feedback, medical blogs, and social media posts. However, sharing sensitive health data with a centralized system for training DL models presents significant privacy, authorization, and regulatory concerns. Federated Learning (FL) provides an efficient framework for local devices (edge devices) to train a shared global model cooperatively and collaboratively while enabling users to control their data locally. In this research, we present a Federated Deep Learning (FDL) framework for estimating users’ emotions in clinical environments that leverages distributed narrative data stored across multiple sources. More specifically, we employ two DL models, including LSTM and Transformers, to train the edge devices using the proposed FDL framework. Nevertheless, the non-IID trait of the subjective text content when used with the FL framework poses many challenges. To simulate this aspect, we employed an efficient Quantity-Level Dirichlet Allocation strategy for distributing data across various clients. For model aggregation, a robust aggregation strategy known as FedMedian is used, which efficiently handles non-IID data. The efficiency and stability of the proposed FDL framework is assessed using two distinct clinical narrative datasets: the Mental Health Status dataset and the Medical Drugs Sentiment Analysis dataset. The empirical study illustrate that the proposed architecture outperforms centralized learning across both the datasets for Emotion Recognition task. Further, we integrated the proposed FDL framework with a Secure Aggregation environment for enhanced security. The study reports that the model’s performance is also consistent with secured aggregation, ensuring data privacy.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Sentiment Analysis and Opinion Mining Emotion and Mood Recognition

Publication details

Journal / source
International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
Pages
1–6
Publication date
26 February 2026
Language
EN