Maribel Acosta presented a tutorial on "Querying Federations of Knowledge Graphs" at the KnowGraphs Winter School 2022.
The Marie Curie ITN-ETN KnowGraphs is a training network that focuses on research about Knowledge Graphs. The Winter School 2022 was held online from 31.01 to 02.02.2022, and comprised 9 tutorials and 3 data challenges.
The tutorial by Maribel Acosta included an overview on the foundations of federated query processing, adaptive techniques for decentralized knowledge graphs, and a framework for query heterogeneous federations of knowledge graphs. The latter is part of a recent publication accepted at The Web Conference 2022, the prime conference for research on the web.
Tutorial Overview
Federations of Knowledge Graphs (KGs) are composed of multiple, decentralized, autonomous sources that expose KGs on the web. To query such KGs, Federated Query Engines (FEQs) implement query processing techniques for reducing the execution time of queries while maximizing the answer completeness.
The first part of this talk will introduce the general architecture of FQEs with an overview of (i) source selection and query decomposition, (ii) query planning, and (iii) query execution techniques. We will learn that due to the decentralized and autonomous aspects of federated KGs, query planning and execution techniques can fail if the runtime conditions are not taken into account. To address this, we will present Adaptive Query Processing (AQP) tailored to federated KGs. We will focus on two types of adaptivity that produce results incrementally and address query performance issues due to network delays or suboptimal plans. In addition, this talk will present a metric for benchmarking querying approaches that produce answers incrementally.
In the second part of this talk, we will concentrate on heterogeneous KG federations. Heterogeneity in query processing can come in different forms, e.g., data models, languages, hardware, interfaces, etc. In this talk, we will focus on heterogeneous federations composed of interfaces with different expressivity. The Linked Data Fragments (LDFs) Framework defines the expressivity of the interfaces in terms of the class of SPARQL expressions they can evaluate and the metadata they provide. This talk will then present the challenges that FQEs face when querying KG federations with heterogeneous LDF interfaces. To address these challenges, in our most recent work, we propose an interface-aware framework that exploits the capabilities of the member of the federations to speed the query execution. This talk will conclude with an outlook on other research problems in the context of federations of KGs.
Relevant Links
Link to the Slides: zenodo.org/record/5972367
Pre-print: arxiv.org/abs/2102.03269