EAFL: Equilibrium Augmentation Mechanism to Enhance Federated Learning for Aspect Category Sentiment Analysis

Khwaja Mutahir Ahmad, Qiao Liu, Abdullah Aman Khan, Yanglei Gan, Run Lin

Abstract: Highlights•Recent advancements in ACSA leveraged PLMs to extract data representations.•Traditional FL encounters difficulties in data heterogeneity or domain-specific tasks.•A novel approach EAFL integrates privacy preservation and augmentation.•A unified model utilize term, polarity, and categorical representations to overcome data heterogeneity.•A parameter-efficient tuning mechanism exploits to fine-tune the model during sentence generation.