Finding optimal surface for hierarchical classification task on an ontology

1. A method, in a data processing system having a processor and a memory, wherein the memory comprises instructions which are executed by the processor to cause the processor to implement a training system for finding an optimal surface for hierarchical classification task on an ontology, the method comprising:

receiving, by the training system, a training data set and a hierarchical ontology data structure;

selecting, by a surface finding component executing within the training system, a current surface that passes through each path from a root to a leaf node in the hierarchical ontology data structure;

determining, by the surface finding component, a plurality of adjacent surfaces that differ from the current surface by one node;

selecting, by the surface finding component, an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage, wherein selecting the optimal surface comprises:

determining a combined coverage and accuracy value for the current surface and for each of the plurality of adjacent surfaces; and

determining whether a combined coverage and accuracy value of a given adjacent surface is greater than a combined coverage and accuracy value of the current surface by a predetermined threshold; and

training, by the training system, a classifier model for a cognitive system using the optimal surface and the training data set.

2. The method of claim 1 , further comprising:

receiving a number of leaf nodes to be covered and a desired accuracy; and

determining whether the optimal surface covers the number of leaf nodes and achieves the desired accuracy.

3. The method of claim 1 , wherein selecting the optimal surface further comprises:

responsive to determining the combined coverage and accuracy value of the given adjacent surface is greater than the combined coverage and accuracy value of the current surface by the predetermined threshold, selecting the given adjacent surface as the current surface and repeating determining a plurality of adjacent surfaces that differ from the current surface by one node and selecting an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage.

4. The method of claim 1 , wherein selecting the optimal surface further comprises:

responsive to determining that there is no given adjacent surface having a combined coverage and accuracy value that is greater than the combined coverage and accuracy value of the current surface, identifying the current surface as the optimal surface.

5. The method of claim 1 , wherein the cognitive system classifies data into classes in the hierarchical ontology data structure under control of the classifier model.

6. The method of claim 5 , wherein the cognitive system classifies to leaf nodes that are not included in the optimal surface separately.

7. The method of claim 1 , wherein training the classifier comprises training on re-labelled training data according to the selected optimal surface and evaluating a resulting cognitive model with a testing dataset.

8. A computer program product comprising a non-transitory computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on at least one processor of a computing device, causes the at least one processor to implement a training system for finding an optimal surface for hierarchical classification task on an ontology, wherein the computer readable program causes the at least one processor to:

receive, by the training system, a training data set and a hierarchical ontology data structure;

select, by a surface finding component executing within the training system, a current surface that passes through each path from a root to a leaf node in the hierarchical ontology data structure;

determine, by the surface finding component, a plurality of adjacent surfaces that differ from the current surface by one node;

select, by the surface finding component, an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage, wherein selecting the optimal surface comprises:

determining a combined coverage and accuracy value for the current surface and for each of the plurality of adjacent surfaces; and

determining whether a combined coverage and accuracy value of a given adjacent surface is greater than a combined coverage and accuracy value of the current surface by a predetermined threshold; and

train, by the training system, a classifier model for a cognitive system using the optimal surface and the training data set.

9. The computer program product of claim 8 , wherein the computer readable program further causes the at least one processor to:

receive a number of leaf nodes to be covered and a desired accuracy; and

determine whether the optimal surface covers the number of leaf nodes and achieves the desired accuracy.

10. The computer program product of claim 8 , wherein selecting the optimal surface further comprises:

responsive to determining the combined coverage and accuracy value of the given adjacent surface is greater than the combined coverage and accuracy value of the current surface by the predetermined threshold, selecting the given adjacent surface as the current surface and repeating determining a plurality of adjacent surfaces that differ from the current surface by one node and selecting an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage.

11. The computer program product of claim 8 , wherein selecting the optimal surface further comprises:

responsive to determining that there is no given adjacent surface having a combined coverage and accuracy value that is greater than the combined coverage and accuracy value of the current surface, identifying the current surface as the optimal surface.

12. The computer program product of claim 8 , wherein the cognitive system classifies data into classes in the hierarchical ontology data structure under control of the classifier model.

13. An apparatus comprising:

at least one processor; and

a memory coupled to the at least one processor, wherein the memory comprises instructions which, when executed by the at least one processor, cause the at least one processor to implement a training system for finding an optimal surface for hierarchical classification task on an ontology, wherein the instructions cause the at least one processor to:

receive, by the training system, a training data set and a hierarchical ontology data structure;

select, by a surface finding component executing within the training system, a current surface that passes through each path from a root to a leaf node in the hierarchical ontology data structure;

determine, by the surface finding component, a plurality of adjacent surfaces that differ from the current surface by one node;

select, by the surface finding component, an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage, wherein selecting the optimal surface comprises:

determining a combined coverage and accuracy value for the current surface and for each of the plurality of adjacent surfaces; and

determining whether a combined coverage and accuracy value of a given adjacent surface is greater than a combined coverage and accuracy value of the current surface by a predetermined threshold; and

train, by the training system, a classifier model for a cognitive system using the optimal surface and the training data set.

14. The apparatus of claim 13 , wherein the instructions further cause the at least one processor to:

receive a number of leaf nodes to be covered and a desired accuracy; and

determine whether the optimal surface covers the number of leaf nodes and achieves the desired accuracy.

15. The apparatus of claim 13 , wherein selecting the optimal surface further comprises:

responsive to determining the combined coverage and accuracy value of the given adjacent surface is greater than the combined coverage and accuracy value of the current surface by the predetermined threshold, selecting the given adjacent surface as the current surface and repeating determining a plurality of adjacent surfaces that differ from the current surface by one node and selecting an optimal surface, based on the current surface and the plurality of adjacent surfaces, that maximizes accuracy and coverage.

16. The apparatus of claim 13 , wherein selecting the optimal surface further comprises:

responsive to determining that there is no given adjacent surface having a combined coverage and accuracy value that is greater than the combined coverage and accuracy value of the current surface, identifying the current surface as the optimal surface.

17. The apparatus of claim 13 , wherein the cognitive system classifies data into classes in the hierarchical ontology data structure under control of the classifier model.

18. The computer program product of claim 8 , wherein the cognitive system classifies data into classes in the hierarchical ontology data structure under control of the classifier model.

19. The computer program product of claim 18 , wherein the cognitive system classifies to leaf nodes that are not included in the optimal surface separately.

20. The apparatus of claim 17 , wherein the cognitive system classifies to leaf nodes that are not included in the optimal surface separately.