Federated Learning Approach towards Sentiment Analysis

• Published in: 2022 2nd International Conference on Technological Advancements in Computational Sciences (ICTACS) pp. 717 - 724
• Main Authors: Bansal, Shefali, Singh, Medha, Bhadauria, Madhulika, Adalakha, Richa
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10.10.2022
• Subjects: Analytical models; Deep Learning; Federated Learning; Heuristic algorithms; Machine Learning; Machine learning algorithms; Natural Language Processing; Prediction algorithms; Predictive models; Sentiment analysis; Training
• DOI: 10.1109/ICTACS56270.2022.9987996

Smartphones have access to a tremendous quantity of data relevant to models, henceforth enhancing the consumer experience. General ML algorithms require centralization of data and models but pose difficulty in GDPR compliance, lack of trust from users' end, and limited transparency. FL is a collective and decentralized approach to Machine Learning that enhances privacy and security of data, complies with GDPR. It also advocates low power consumption and latency. This paper delivers the application of Federated Learning in sentiment analysis, in comparison to the Deep Learning algorithms - RNN, CNN used. The sentiments of the text are segmented in accordance with the spectrum of emotions they articulate - positive, negative, and neutral using Deep Learning algorithms. We have also added a web component (a dynamic web-app) to the sentiment analysis model to automate the prediction process.