Vers une Web des données en Lecture/Écriture
Ibáñez, Luis-Daniel · HAL (Le Centre pour la Communication Scientifique Directe) · 2015
The Linked Data initiative has made available millions of pieces of data for querying through a federation of autonomous participants. However, the Web of Linked data suffers of problems of data heterogeneity and quality. We cast the problem of integrating heterogeneous data sources as a Local-as-View mediation (LAV) problem, unfortunately, LAV may require the execution of a number of “rewritings” exponential on the number of query subgoals. Wepropose the Graph-Union (GUN) strategy to maximise the results obtained from a subset of rewritings. Compared to traditional rewriting execution strategies, GUN improves execution time andnumber of results obtained in exchange of higher memory consumption. Once data can be queried data consumers can detect quality issues, but to resolve them they need to write on the data ofthe sources, i.e., to evolve Linked Data from Read/Only to Read-Write. However, writing among autonomous participantsraises consistency issues. We model the Read-Write Linked Data as a social network where actors copy the data they are interested into, update it and publish updates to exchange with others. We propose two algorithms for update exchange: SU-Set, that achieves Strong Eventual Consistency (SEC) and Col-Graph, that achievesFragment Consistency, stronger than SEC. We analyze the worst and best case complexities of both algorithms and estimate experimentally the average complexity of Col-Graph, resultssuggest that is feasible for social network topologies.