Enabling Web-Scale Knowledge Graphs Querying

Abstract

Knowledge Graphs (KGs) have become an asset for integrating and consuming data from heterogeneous sources. KGs have an influence on several domains such as health-care, manufacturing, transportation and energy. Over the years, the Web of Data has grown significantly. Today, answering complex queries on open KGs is practically impossible due to the SPARQL endpoints availability problem caused by well-known scalability and load balancing issues when hosting Web-size data for concurrent clients. To maintain reliable and responsive open Knowledge Graph query services, several solutions have been proposed: while SPARQL endpoints enforce restrictions on server usage such as imposing limited query execution time or providing partial query results, alternative solutions such as Triple Pattern Fragments (TPF) attempts to tackle the problem of availability by pushing query processing workload to the client-side but suffer from the unnecessary transfer of irrelevant data on complex queries as a result of the large intermediate results. The aim of our research is to develop a new generation of smart clients and servers to balance the load between servers and clients, with the best possible query execution performance, and at the same time reducing data transfer volume, by combining SPARQL endpoints, TPF and shipping compressed KG partitions. The proposed solution shall, on the server-side, offer a suitable query execution service according to the current status of the server workload. On the client-side, we plan research on novel client-side caching mechanisms on the basis of compressed and queryable KG partitions that can be distributed in a modular fashion. In addition, we plan to leverage query logs to optimize the number and the distribution of partitions as well as distributing the query load across a network of collaborative clients.