A personal perspective on keyword search over data graphs

Theoretical and practical issues pertaining to keyword search over data graphs are discussed. A formal model and algorithms for enumerating answers (by operating directly on the data graph) are described. Various aspects of a system are explained, including the object-connector-property data model, how it is used to construct a data graph from an XML document, how to deal with redundancies in the source data, what are duplicate answers, implementation and GUI. An approach to ranking that combines textual relevance with semantic considerations is described. It is argued that search over data graphs is inherently a two-dimensional process, where the goal is not just to find particular content but also to collect information on how the desired data may be semantically connected.