Beyond Hypertext: A Web of Dynamic, Structured Data Sources
Abstract
The Web has been an incredibly effective means of disseminating information,
primarily in the form of interlinked HTML documents. Unfortunately this Web
is geared towards presenting information to users -- not towards
facilitating data flow among the producers, curators, and integrators of
data. In this talk I describe our ongoing efforts to design a global-scale
"Web of structured data and data sources" where data can flow along links.
Our work revolves around the Orchestra system, which provides a means of
linking data sources, such that data and updates may be translated and
propagated along the links. Some sources may contain incompatible or
inconsistent data -- not only due to variations in data freshness, but also
due to different *beliefs* among sources. Such conflicts must be managed by
considering freshness and authoritativeness. Orchestra develops techniques
for tracking data usage and derivation (provenance), and for using
provenance and the authoritativeness of the sources to help manage
consistency. Orchestra's query layer, Q, learns about authoritativeness,
given user feedback about the relevance of query answers.
I will wrap up the talk by briefly describing our ongoing work to build the
successor to Orchestra, Concerto, which *makes recommendations* about data
and sources likely to be of interest to the user, given the overall data
usage patterns in the system.
This is joint work with Todd Green, Grigoris Karvounarakis, Nicholas Taylor,
Partha Pratim Talukdar, Marie Jacob, Val Tannen, Fernando Pereira, and
Sudipto Guha.
|
