A recommendation system plays an indispensable part in real-world applications to suggest relevant items for the interest of the users. In contemplating to imply better recommendations to the user, the recommendation system gathers user-s private data such as the behavioral and contextual information, the domain knowledge, the item metadata, the purchase history, the recommendation feedback, the social data, and also utilizes multiple data sources from other organizations. All the collected information are stored in a centralized server which leads to cause the data privacy issue in the recommendation system. As public awareness around data and privacy laws increases, appreciating data privacy in recommendation systems will be a sustainable consideration. Privacy-preserving recommendations are an essential benefit in the field of healthcare and finance.
Federated learning is a new algorithmic approach that trains the learning models based on data centralization, which provides better data protection. Federated learning trains the recommendation models cooperatively in a secure manner while keeping the user’s private data locally. Challenges and future directions of federated learning in recommendation systems are federated deep model and graph model for the recommendation, recommendation based on reinforcement learning, recall and ranking based federated recommendation system, cooperation of malicious participants in federated recommendation system, a non-IID data-based federated recommendation system and improvement in communication cost, flexibility, scalability of federated recommendation system.
• Federated learning leverages the cooperative training of machine learning and deep learning methods for complex edge network optimization with heterogeneous devices.
• Challenges of data silos and data sensibility are overcome by collaboratively decentralized privacy-preserving technology - Federated learning.
• Emerging advances in federated learning are federated recommendation systems, which significantly aim to construct high-performance recommendation systems by connecting data repositories without exasperating data security and privacy.
• A robust learning strategy is instigated in federated recommendation systems to empower resistance against low-cost poisoning attacks conducive to improving the performance of the model.
• Federated recommendations facilitate recommendations for both public users and private users, which allows for building privacy-preserving federated learning for page recommendations.
• News recommendation framework with privacy protection is achieved by federated learning, without centralized storage of useful information, and yields accurate news recommendations.
• Recently, in order to limit the communication cost, and improvised many objective federated recommendation models with a novel parameter reduction strategy have been developed.
• Also, to address the issue of non-independent and identically distribution in federated learning, a personalized federated recommender based on the clustering of historical parameters is evolved.
• Despite the fact of decentralized recommendation in federated recommendation systems, there are possible complexities in terms of both algorithmic and systemic levels that are to be addressed in the future for implementing federated recommenders.
