An opinion mining with federated learning on the Afghan-People survey dataset

Abstract

Internet and digital devices have caused a flood of information in the fourth industrial revolution era. Health-care, government, banks, military, and the banking sector, have embraced machine learning as a convenient way to recognize patterns in data. For a conventional machine learning model, the information from the data owner must be uploaded in a centralized location for training, which causes data owners to worry about the lack of their private information assurance. On the other hand, a massive amount of computing power is available to train intelligent models. In this way, federated learning becomes more and more popular. It preserves privacy of data owners and allows models to be trained over distributed clients. In this work, the federated learning approach is implemented using a neural network to model several Afghan-People opinion investigations such as the question about country's overall situation or people's security and safety in a distributed computing environment with measurable results. Further, the performance of federated learning using the same dataset is compared with centralized machine learning.