Semantic Network with Selective Indexing

• US 20110106745A1
• Filed: 11/03/2009
• Published: 05/05/2011
• Est. Priority Date: 11/03/2009
• Status: Active Grant

• Current Assignee: Postq IPR OY
• Original Assignee: TATU Ylonen OY Limited
• Inventors: Ylonen, Tatu J.
• Granted Patent: US 8,285,664 B2
• Time in Patent Office: Days
• US Class Current: 706/47
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

Indexing overhead in a semantic network is reduced by omitting some links from indexes in semantic network elements in response to one or more omission rules. Inference strategy over the semantic network is modified to prioritize inference strategies in such a way that the impact of the omissions on inference results is minimized.

Claims

1. A computer comprising a semantic network stored in its accessible memory, said semantic network comprising semantic network elements, the computer comprising:

a first omission rule means; and

a first index maintainer responsive to the first omission rule means, the omission rule means causing the first index maintainer to omit at least one link referencing a first node from a first index associated with the first node.

2. The computer of claim 1, further comprising:

an inference system responsive to the first omission rule means, the first omission rule means causing the inference system to prioritize an inference strategy that avoids trying to follow, from the first node, links that have been omitted from the first index, over an inference strategy that tries to follow such links.

3. The computer of claim 1, wherein:

links can reference other links;

a second link has a second index associated with it, the second index comprising references to the second link by other links; and

the second index is maintained by a second index maintainer, which is responsive to a second omission rule means, the second omission rule means causing the second index maintainer to omit at least one link referencing the second link from the second index.

4. The computer of claim 1, wherein the index maintainer implements omitting at least one link by dropping the link from the first index when signalled to do so by a second omission rule.

5. The computer of claim 4, wherein the selection of the link to drop is responsive to the relative prominence of the links in the first index.

6. The computer of claim 4, wherein a link dropped from the first index is reclaimed by the garbage collector if it is no longer present in any index and no other link references it.

7. The computer of claim 1, wherein the first omission rule is responsive to the argument position in the link that references the associated node.

8. The computer of claim 1, wherein the omission rule is responsive to statistics collected about use of links in inference.

9. The computer of claim 1, further comprising an omission rule learner configured to modify at least one omission rule.

10. A method of maintaining an index in a semantic network element, comprising:

adding, by a computer, a plurality of links into an index associated with a semantic network element; and

omitting, by the computer, at least one link from the index based on a first omission rule.

11. The method of claim 10, further comprising:

making inferences over the semantic network, prioritizing an inference strategy that avoids trying to follow links from indexes from which some links have been omitted over a strategy that tries to follow links from such indexes.

12. The method of claim 10, wherein omitting at least one link is implemented by dropping it from the index when signalled to do so by a second omission rule.

13. The method of claim 10, further comprising:

modifying at least one omission rule based on information collected about the use of indexes in inferences.

14. A computer program product stored on a computer readable medium, the product operable to maintain an index in a semantic network element, and the product comprising:

a computer readable program code means for causing a computer to index a plurality of links into an index associated with a semantic network element; and

a computer readable program code means for causing the computer to omit at least one link from the index based on a first omission rule.

15. The computer program product of claim 14, further comprising:

a computer readable program code means for causing the computer to make inferences over the semantic network, prioritizing an inference strategy that avoids trying to follow links from indexes from which some links have been omitted over a strategy that tries to follow links from such indexes.

16. The computer program product of claim 14, further comprising:

a computer readable program code means for causing the computer to implement omitting at least one link by dropping it from the index when signalled to do so by a second omission rule.

17. The computer program product of claim 14, further comprising:

a computer readable program code means for causing the computer to modify at least one omission rule based on information collected about the use of indexes in inferences.