Optimizing Collaborative Filtering In Federated Learning By Aligning The Model With Different Data Partitions

In the context of recommendation systems, federated learning is a promising technique that, through decentralization, benefits from different client con-tributions. Nevertheless, the locally employed model should maximize the federated results, something that may be achieved by taking into considera-tion the data partition present in each client. Therefore, by aligning the cho-sen model with the federated data partition in the local clients, or vice ver-sa, it will be possible to present more accurate recommendations to the cli-ents, enhancing, therefore, recommendation quality. This paper intends to explore which data partition suits collaborative filtering best, by perform-ing a practical implementation of this model in a federated scenario, with a different number of clients and different data partition types.