Structural similarity-based federated learning for mining OLAP analytical cubes in distributed environments

The rising prevalence of distributed data management systems and the escalating trend of utilizing privacy-preserving data management systems render OLAP-based data analytics on dispersed and heterogeneous data sources a significant problem. The traditional method of OLAP cube creation necessitates centralized data aggregation, which is impractical in distributed data management systems due to concerns over data privacy, communication, and ownership. This research proposes a novel framework termed Federated Structural Similarity Learning for OLAP cube extraction that does not require the sharing of transactional data among disparate sources. The primary concept of the proposed methodology is to acquire local structural representations and OLAP-based analytical structural similarity among various sources through a federated learning framework, which preserves the structural relationships between disparate data sources during OLAP-based data analysis. The proposed methodology is evaluated using the Online Retail II dataset, and its efficacy is assessed by juxtaposing its performance against two prominent methodologies, specifically PACFL and FedSim, based on OLAP structural representation, analytical structural similarity, federated convergence time, and federated communication cost. The proposed method enhances OLAP structural representation by ۱۳%, analytical structural similarity by ۱۵%, federated convergence time by ۲۵%, and federated communication cost by ۳۲%, demonstrating that federated structural similarity learning is an effective, efficient, and viable strategy for distributed OLAP-based data analytics in practical scenarios.

کلیدواژه ها:

OLAP cubes ، multidimensional analysis ، distributed data ، data heterogeneity