The Web of Data is partially unavailable, so what?
Katja Hose · 2020
More and more knowledge graphs are becoming available on the Web, many of them accessible via queryable interfaces, such as SPARQL endpoints. These knowledge graphs are often linked to each other and in doing so form the so-called Web of Data. The great strength and potential of exploiting these links is that we can obtain answers for queries that would not yield any results over the knowledge graphs if considered individually. However, one of the fundamental problems is that Web-accessible knowledge graphs are not always available when we would like to access them. This means that the exact same query issued at different points in time might have different results or, in the worst case, no results if one of the knowledge graphs that are essential to answer the query is (temporarily) not available. Often this is caused by crashes of the server hosting the knowledge graph - a problem that is deeply rooted within the architecture of the current Web of Data, i.e., relying on data providers to keep interfaces and knowledge graphs available and accessible. Building upon the basics of federated query processing over knowledge graphs, this talk will discuss challenges as well as strategies on how to overcome the unavailability problem and keep the data available and queryable.