Integrating (Very) Heterogeneous Data Sources: A Structured and an Unstructured Perspective
Résumé
Data integration is a broad area of data management research. It has lead to the development of many useful tools and concepts, each appropriate in a certain class of applicative settings.
We consider the setting in which data sources have heterogeneous data models. This setting is of increasing relevance, as the (once predominant) relational databases are supplemented by data exchanged in formats such as JSON or XML, graphs such as Linked Open (RDF) data, or matrix (numerical) etc. We describe two lines of work in this setting. The first aims on improving performance in a polystore setting, where data sources are queried through a structure, composite query language; the focus here is on dramatically improving performance through the use of view-based rewriting techniques. The second data integration setting assumes that sources are much too heterogeneous for structured querying and thus, explore keyword-based search in an integrated graph built from all the available data.
Designing and setting up data integration architectures remains a rather complex task; data heterogeneity makes it all the more challenging. We believe much remains to be done to consolidate and advance in this area in the future.