Decentralized Learning with Average Difference Aggregation for Proactive Online Social Care

The Internet and the Web are being increasingly used in proactive social care to provide people, especially the vulnerable, with a better life and services, and their derived social services generate enormous data. However, privacy concerns require the strict protection of a user's data which leads to a dilemma between pursuing better intelligent services and privacy preservation. To solve this dilemma, this paper develops a decentralized learning framework to enable proactive social care equipped with artificial intelligence and data protection. The proposed learning framework trains a secure local model for each user using their own datasets. Each user's device only shares its local model's parameters with the central model server without exposing any data, and the server integrates multiple users' models to learn a global data-free model. To improve the generalizability of the global model, we further propose a novel model aggregation algorithm, namely the average difference descent aggregation (AvgDiffAgg for short). In particular, to evaluate the effectiveness of the learning algorithm, we use a case study on the early detection and prevention of suicidal ideation, and the experiment results on four datasets derived from social media demonstrate the effectiveness of the proposed learning method.