Physical knowledge graph design for efficient federated query processing
Julián Arenas Guerrero · 2020
In the current context of data prevalence, knowledge graphs have emerged as a popular method to integrate and extract value of heterogeneous large-scale data sources. In this thesis, we propose a novel idea for efficiently computing queries in a knowledge graph: the optimization of its physical design by combining materialization and virtualization. To this end, we carried out an empirical comparison between these two approaches and studied the parameters that have an impact on their performances. State of the art benchmarks and engines are used throughout the work. We found that, for a lower number of triples, the efficiency is enhanced in the case of materialization. In contrast, virtualization tends to improve as the data size increases. The viability of our approach is confirmed, proving that this hybrid approach can indeed enhance considerably query performance by materializing specific parts of knowledge graphs. It is, to our knowledge, the first time that this idea has been tested, and we believe that this work can be the starting point for future research focusing in query performance.