Data Integration and Data Exchange: It's Really About Time

With the deluge in the amount and variety of data in the world, it is rare for data that describes an entity to be completely contained and managed by a single data source. As a consequence, there is often great value in combining data about an entity from multiple sources, and also from versions of data reported by the same source over time. Data integration in which multiple dimensions of time may be expressed explicitly (e.g., as part of the data itself) or implicitly (e.g., the publication date of a data source), must be performed with great care. This is because each data source contains only partial (time-specific) knowledge about an entity, and thus their collective knowledge about the entity may contain conflicts that need to be resolved. In this paper, we call for a formal framework for data integration and data exchange across time that would facilitate the creation of consistent and integrated longitudinal knowledge about entities. We call such longitudinal knowledge of an entity its whenprovenance, which intuitively corresponds to when one knows what one knows about the entity. We believe that the vision and research directions described in this paper will serve to instigate the research and development of the next generation data integration and data exchange system, where both data and time can be reasoned on equal footing.