Queries over Virtual Nested Objects

We describe a system that allows the easy specification and efficient support of queries specified as a set of attribute/ predicate/value triplets over virtual nested objects constructed from relational databases. For example, the Internet Movie Database (imdb) provides virtual nested objects such as ”movies” (each one containing a movie tuple, multiple actor tuples and more) and ”actors” (containing multiple movies). Queries may contain both boolean and text/fuzzy predicates and may be directed to one or more virtual nested object sets. We define ranked query semantics that capture the common requirements that (1) individual tuples within a nested object may only satisfy a subset of all the predicates and (2) an object that contains a tuple that satisfies more than one predicates should (all other things being equal) be ranked higher than an object where such predicates are satisfied over multiple tuples. The system fully utilizes the existing indices in the relational databases and combines special purpose algorithms with database accesses. Experiments demonstrate that the obtained performance is significantly better than the performance obtained by fully deferring query evaluation to SQL queries.