Jabba-type override for correcting or improving output of a model

• US 9,195,940 B2
• Filed: 03/15/2013
• Issued: 11/24/2015
• Est. Priority Date: 03/15/2013
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Coalson, Josh
• Primary Examiner(s): Holmes, Michael B
• Application Number:
  US13/844,013
• Publication Number:
  US 20140279743A1
• Time in Patent Office: 984 Days
• Field of Search: 706/12
• US Class Current: 1/1
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

Example methods, apparatuses, or articles of manufacture are disclosed that may be implemented, in whole or in part, using one or more computing devices to facilitate or otherwise support one or more processes or operations for a Jabba-type override for correcting or improving output of a model, such as a machine-learned model, for example.

Claims

1. A method comprising:

electronically accessing one or more Jabba rules descriptive of one or more named terms characterizing a preferred output of a model, said one or more named terms corresponding to at least one of the following;

a match expression;

a sequence;

an alternation;

a call;

or any combination thereof; and

generating one or more digital signals representative of an override path by determining a machine for said model based, at least in part, on one or more sequences of candidate atoms and said one or more named terms;

wherein said machine comprises a Jabba-type machine capable of executing one or more actions on nodes of an associated graph representing match points along said one or more sequences of said candidate atoms.

2. The method of claim 1, and further comprising generating an override output of said model by augmenting said preferred output of said model with said one or more sequences of said candidate atoms based, at least in part, on an interpretation score determined for said override output.

3. The method of claim 1, and further comprising generating an override output of said model by merging said preferred output of said model with said one or more sequences of said candidate atoms based, at least in part, on an interpretation score determined for said override output.

4. The method of claim 1, wherein said override path is generated selectively in response to a search query.

5. The method of claim 1, wherein said machine comprises a nondeterministic finite state automaton (NDFA).

6. The method of claim 1, wherein said machine comprises an override Jabba-type machine determined before an override output of said model is generated.

7. The method of claim 1, wherein said preferred output of said model comprises an override output of said model.

8. The method of claim 1, wherein said one or more sequences of said candidate atoms are associated with a search query.

9. The method of claim 1, wherein said one or more Jabba rules are implemented via at least one of the following:

editorial input;

user feedback;

a feed of structured data;

or any combination thereof.

10. The method of claim 1, wherein said model comprises a machine-learned model.

11. The method of claim 1, wherein said preferred output comprises a known output.

12. The method of claim 11, wherein said known output is determined by a human editor.

13. An apparatus comprising:

a computing platform comprising at least one processor to;

electronically access one or more Jabba rules to be descriptive of one or more named terms to characterize a preferred output of a model, wherein said one or more named terms to correspond to at least one of the following;

a match expression;

a sequence;

an alternation;

a call;

or any combination thereof; and

generate one or more digital signals to be representative of an override path to determine a machine to be associated with said model to be based, at least in part, on one or more sequences of candidate atoms and said one or more named terms;

said machine to comprise a Jabba-type machine to be capable of execution of one or more actions on nodes of an associated graph to represent match points along said one or more sequences of said candidate atoms.

14. The apparatus of claim 13, wherein said at least one processor further to generate an override output of said model to augment said preferred output of said model with said one or more sequences of said candidate atoms to be based, at least in part, on an interpretation score to be determined for said override output.

15. The apparatus of claim 13, wherein said at least one processor further to generate an override output of said model to merge said preferred output of said model with said one or more sequences of said candidate atoms to be based, at least in part, on an interpretation score to be determined for said override output.

16. The apparatus of claim 13, wherein said machine to comprise an override Jabba-type machine to be determined before an override output of said model is generated.

17. The apparatus of claim 13, wherein said preferred output of said model to comprise an override output of said model.

18. An article comprising:

a non-transitory storage medium having instructions stored thereon executable by a special purpose computing platform to;

electronically access one or more Jabba rules to be descriptive of one or more named terms to characterize a preferred output of a model, wherein said one or more named terms to correspond to at least one of the following;

a match expression;

a sequence;

an alternation;

a call;

or any combination thereof; and

generate one or more digital signals to be representative of an override path to determine a machine to be associated with said model to be based, at least in part, on one or more sequences of candidate atoms and said one or more named terms;

said non-transitory storage medium having further instructions so that said machine is to comprise a Jabba-type machine to be capable of execution of one or more actions on nodes of an associated graph to represent match points along said one or more sequences of said candidate atoms.

19. The article of claim 18, wherein said non-transitory storage medium further includes instructions to generate an override output of said model to augment said preferred output of said model with said one or more sequences of said candidate atoms to be based, at least in part, on an interpretation score to be determined for said override output.

20. The article of claim 18, wherein said non-transitory storage medium further includes instructions to generate an override output of said model to merge said preferred output of said model with said one or more sequences of said candidate atoms to be based, at least in part, on an interpretation score to be determined for said override output.