Graph search system and method for querying loosely integrated data

1. A method of executing a query on linked data sources comprising:

generating an instance graph expressing relationships between objects in said linked data sources;

receiving a query including at least two terms;

rewriting, heuristically, the query into an ordered list of sub-query terms, wherein the ordered list begins with first and second search terms;

executing, with a processor, said first search term on said instance graph, wherein the execution outputs an execution result subset and wherein said executing said first search term further comprises performing a relationship search that ranks each object in said instance graph with respect to said execution result subset;

generating a summary graph using the execution result subset, wherein objects having a score below a predetermined threshold are filtered out of said summary graph during said relationship search;

aggregating to said summary graph two or more of said filtered out objects which are related to at least one object having a score above said predetermined threshold; and

executing said second search term on said summary graph.

2. The method of claim 1 wherein said relationship search uses a fuzzy search operator to derive said summary graph and wherein said summary graph indicates how closely related an object in said linked data sources is to said first and second search terms.

3. The method of claim 1 wherein said relationship search uses a random walk to derive said summary graph and wherein said summary graph indicates how closely related an object in said linked data sources is to said first and second search terms.

4. The method of claim 1 wherein said relationship search uses an ObjectRank method to derive said summary graph and wherein said summary graph indicates how closely related an object in said linked data sources is to said first and second search terms.

5. The method of claim 1 further comprising generating a presentation graph from said summary graph that visually displays clusters of related objects in said linked data sources.

6. A method of finding relationships between objects in a database comprising:

generating an instance graph expressing relationships between the objects in said database;

receiving a query including at least two terms;

rewriting, heuristically, the query into an ordered list of sub-query terms, wherein the ordered list begins with first and second search terms;

executing, with a processor, the first search term in a query, wherein said executing derives a subset of said database;

performing a relationship search that ranks each object in said instance graph with respect to said subset;

filtering out the objects of said relationship search, wherein each filtered out object has a score below a predetermined threshold;

generating a summary graph using the subset of said executing;

aggregating to said summary graph two or more of said filtered out objects which are related to at least one object having a score above said predetermined threshold; and

executing said second search term on said summary graph, wherein the execution outputs second subset.

7. The method of claim 6 wherein said received query includes a third search term and further comprising executing said third search term on the results of said second subset.

8. The method of claim 6 wherein said performing a relationship search comprises performing a random walk.

9. The method of claim 6 wherein said performing a relationship search comprises using an ObjectRank method.

10. The method of claim 6 further comprising generating a presentation graph from said summary graph that visually displays clusters of related objects in said linked databases.

11. A computer program product comprising a computer readable storage medium having a computer readable program, wherein said computer readable program when executed on a computer causes said computer to:

generate an instance graph expressing relationships between objects in linked data sources;

receive a query including at least two terms;

rewrite, heuristically, the query into an ordered list of sub-query terms, wherein the ordered list begins with first and second search terms;

execute said first search term on said instance graph, wherein the executing outputs execution result subset and wherein, said executing further comprises performing a relationship search that ranks each object in said instance graph with respect to said execution result subset;

filter out the objects of said relationship search, wherein each filtered out object has a score below a predetermined threshold;

generate a summary graph using the execution result subset;

aggregate to said summary graph two or more of said filtered out objects which are related to at least one object having a score above said predetermined threshold; and

execute said second search term on said summary graph.