Transcript Bruce Croft

DBMS with probabilistic model
 Probabilities and uncertainty taken seriously as
foundation for the database model – not an addon
 old idea but never taken seriously
 e.g. Fuhr’s Probabilistic Datalog
 Probabilistic retrieval model, probabilistic
schema, uncertain queries, learning, appropriate
data structures, etc.
IR queries for DBMS
 Not the same as NL queries work done in 80s
 Take a NL query, apply it to a DBMS, determine
what type of answer can be provided
 may be fact-type answer (QA)
 may be ranked tuples (IR)
 Useful for distributed, heterogeneous search
 e.g. Hidden Web
 should not require wrappers