Automatic combination and mapping of text-mining services
View Patent ↗Embodiments of systems and methods for automatic combination of text mining services, may comprise an instance generation component and an auto-mapping component. From common text sources, the instance generation component generates instances for taxonomy elements attached to a particular text mining service. These instances are then forwarded to the auto-mapping component, which computes a mapping between different taxonomies. This mapping may be saved to avoid repeated instance generation and mapping processes for similar taxonomies. The computed mapping may in turn be input to a result combiner element, together with extraction results from calling the different text mining services. The result combiner returns the merged result sets to the execution, and finally to the user or API.
1. A computer-implemented method of combining text mining services that rely upon different taxonomies, the method comprising:
applying a first text mining service relying upon a first taxonomy, to a knowledge base comprising a plurality of documents according to a first iterative instance generation process to generate a plurality of instances of an entity type of the first taxonomy;
applying a second text mining service relying upon a second taxonomy, to the knowledge base according to a second iterative instance generation process to generate a plurality of instances of an entity type of the second taxonomy;
utilizing the plurality of instances of the first taxonomy and the plurality of instances of the second taxonomy to construct an instance-matching procedure;
applying the first text mining service to a target text document smaller than the knowledge base to produce a plurality of first taxonomy entity types;
applying the second text mining service to the target text document to produce a plurality of second taxonomy entity types;
creating a mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the instance-matching procedure;
using the mapping to create a merged result that compares a result of applying the first text mining service to the target text document, with a result of applying the second text mining service to the target text document; and
displaying the merged result to a user.
2. The method of claim 1 wherein:
the first taxonomy comprises a first metadata and the second taxonomy comprises a second metadata; and
creating the mapping further comprises comparing the first metadata and the second metadata.
3. The method of claim 2 wherein the first metadata comprises a first taxonomy entity type name, a first taxonomy entity type description, or a first taxonomy entity type location within a first taxonomy hierarchy.
4. The method of claim 1 further comprising:
applying the first text mining service to a broadened common knowledge base that is larger than the knowledge base, in order to generate an extended plurality of instances of the entity type of the first taxonomy;
applying the second text mining service to the broadened common knowledge base to generate an extended plurality of instances of the entity type of the second taxonomy;
utilizing the extended plurality of instances of the first taxonomy and the extended plurality of instances of the second taxonomy to construct a refined instance-matching procedure; and
creating the mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the refined instance-matching procedure.
5. The method of claim 1 wherein:
the plurality of first taxonomy entity types comprises a first quantity (S);
the plurality of the second taxonomy entity types comprises a second quantity (T); and
an output of the instance-matching procedure comprises a similarity matrix comprising S*T entries, each entry reflecting a similarity between a first taxonomy entity type and a second taxonomy entity type.
6. The method of claim 1 further comprising using a mapping rewriting technique to revise the mapping.
7. A non-transitory computer readable storage medium embodying a computer program for performing a method of combining text mining services that rely upon different taxonomies, said method comprising:
applying a first text mining service relying upon a first taxonomy, to a knowledge base comprising a plurality of documents according to a first iterative instance generation process in order to generate a plurality of instances of an entity type of the first taxonomy;
applying a second text mining service relying upon a second taxonomy, to the knowledge base according to a second iterative instance generation process to generate a plurality of instances of an entity type of the second taxonomy;
utilizing the plurality of instances of the first taxonomy and the plurality of instances of the second taxonomy to construct an instance-matching procedure;
applying the first text mining service to a target text document smaller than the knowledge base to produce a plurality of first taxonomy entity types;
applying the second text mining service to the target text document to produce a plurality of second taxonomy entity types;
creating a mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the instance-matching procedure;
using the mapping to create a merged result that compares a result of applying the first text mining service to the target text document, with a result of applying the second text mining service to the target text document; and
displaying the merged result to a user.
8. The non-transitory computer readable storage medium of claim 7 wherein:
the first taxonomy comprises a first metadata and the second taxonomy comprises a second metadata; and
creating the mapping further comprises comparing the first metadata and the second metadata.
9. The non-transitory computer readable storage medium of claim 8 wherein the first metadata comprises a first taxonomy entity type name, a first taxonomy entity type description, or a first taxonomy entity type location within a first taxonomy hierarchy.
10. The non-transitory computer readable storage medium of claim 7 wherein the method further comprises:
applying the first text mining service to a broadened common knowledge base that is larger than the knowledge base, in order to generate an extended plurality of instances of the entity type of the first taxonomy;
applying the second text mining service to the broadened common knowledge base to generate an extended plurality of instances of the entity type of the second taxonomy;
utilizing the extended plurality of instances of the first taxonomy and the extended plurality of instances of the second taxonomy to construct a refined instance-matching procedure; and
creating the mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the refined instance-matching procedure.
11. The non-transitory computer readable storage medium of claim 7 wherein:
the plurality of first taxonomy entity types comprises a first quantity (S);
the plurality of the second taxonomy entity types comprises a second quantity (T); and
an output of the instance-matching procedure comprises a similarity matrix comprising S*T entries, each entry reflecting a similarity between a first taxonomy entity type and a second taxonomy entity type.
12. The non-transitory computer readable storage medium of claim 7 wherein the method further comprises using a mapping rewriting technique to revise the mapping.
13. A computer system configured to combine text mining services that rely upon different taxonomies, the computer system comprising:
one or more processors;
a software program, executable on said computer system, the software program configured to:
apply a first text mining service relying upon a first taxonomy, to a knowledge base comprising a plurality of documents according to a first iterative instance generation process in order to generate a plurality of instances of an entity type of the first taxonomy;
apply a second text mining service relying upon a second taxonomy, to the knowledge base according to a second iterative instance generation process to generate a plurality of instances of an entity type of the second taxonomy;
utilize the plurality of instances of the first taxonomy and the plurality of instances of the second taxonomy to construct an instance-matching procedure;
apply the first text mining service to a target text document smaller than the knowledge base to produce a plurality of first taxonomy entity types;
apply the second text mining service to the target text document to produce a plurality of second taxonomy entity types;
create a mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the instance-matching procedure;
use the mapping to create a merged result that compares a result of applying the first text mining service to the target text document, with a result of applying the second text mining service to the target text document; and
display the merged result to a user.
14. The computer system of claim 13 wherein:
the first taxonomy comprises a first metadata and the second taxonomy comprises a second metadata; and
creating the mapping further comprises comparing the first metadata and the second metadata.
15. The computer system of claim 14 wherein the first metadata comprises a first taxonomy entity type name, a first taxonomy entity type description, or a first taxonomy entity type location within a first taxonomy hierarchy.
16. The computer system of claim 13 further comprising:
applying the first text mining service to a broadened common knowledge base that is larger than the knowledge base, in order to generate an extended plurality of instances of the entity type of the first taxonomy;
applying the second text mining service to the broadened common knowledge base to generate an extended plurality of instances of the entity type of the second taxonomy;
utilizing the extended plurality of instances of the first taxonomy and the extended plurality of instances of the second taxonomy to construct a refined instance-matching procedure; and
creating the mapping of the plurality of first taxonomy entity types to the plurality of the second taxonomy entity types based at least in part on the refined instance-matching procedure.
17. The computer system of claim 13 wherein:
the plurality of first taxonomy entity types comprises a first quantity (S);
the plurality of the second taxonomy entity types comprises a second quantity (T); and
an output of the instance-matching procedure comprises a similarity matrix comprising S*T entries, each entry reflecting a similarity between a first taxonomy entity type and a second taxonomy entity type.
18. The computer system of claim 13 further comprising using a mapping rewriting technique to revise the mapping.