Col-Graph: Towards Writable and Scalable Linked Open Data

摘要

Linked Open Data faces severe issues of scalability, availability and data quality. These issues are observed by data consumers performing federated queries; SPARQL endpoints do not respond and results can be wrong or out-of-date. If a data consumer finds an error, how can she fix it? This raises the issue of the writability of Linked Data. In this paper, we devise an extension of the federation of Linked Data to data consumers. A data consumer can make partial copies of different datasets and make them available through a SPARQL endpoint. A data consumer can update her local copy and share updates with data providers and consumers. Update sharing improves general data quality, and replicated data creates opportunities for federated query engines to improve availability. However, when updates occur in an uncontrolled way, consistency issues arise. In this paper, we define fragments as SPARQL CONSTRUCT federated queries and propose a correction criterion to maintain these fragments incrementally without reevaluating the query. We define a coordination free protocol based on the counting of triples derivations and provenance. We analyze the theoretical complexity of the protocol in time, space and traffic. Experimental results suggest the scalability of our approach to Linked Data.