Type-enabled Keyword Searches with Uncertain Schema

Soumen Chakrabarti
Department of Computer Science, Indian Institute of Technology, Bombay, India


Web search is beginning to exploit powerful machine learning tools
that annotate the corpus with entities and relationships. Such annotations, to-
gether with techniques for disambiguation and linkage resolution, will lead to
graphical models that capture °exible type information, as well as represent the
inherent uncertainty in the extracted structure. The next piece in the puzzle is
a schema-agnostic query language that enables embedding type constraints in
a user-friendly way; alternatively, machine learning techniques can extract type
specs from unstructured queries. The ¯nal challenge is to devise a model for
matching, scoring, and top-k search that naturally handles the uncertainty in
the graph structure, and leads to manageable indices, scalable query execution
algorithms, and user satisfaction.