Extracting data

1. A system for extracting data, comprising:

a processor;

a storage system, comprising:

an unstructured document;

a rule comprising a plurality of prefixes for a target entity and a plurality of suffixes for the target entity, and

code configured to direct the processor to:

perform an entity recognition on the unstructured document to mark entities by type;

apply the rule to marked entities; and

generate a confidence value for each of the marked entities, where the confidence value indicates a degree of match to the rule.

2. The system of claim 1 , comprising code configured to direct the processor to select the entity with a highest confidence value as an extraction target.

3. The system of claim 1 , comprising code configured to direct the processor to annotate each token of the unstructured document with a parts-of-speech tag.

4. The system of claim 3 , wherein the memory comprises code configured to direct the processor to direct the processor to annotate each token of the unstructured document with a parts-of-speech tag.

5. The system of claim 1 , wherein the memory comprises code configured to direct the processor to:

obtain a plurality of training documents, wherein an entity of interest is marked in each of the plurality of training documents;

generate a plurality of prefixes for the entity of interest;

generate a plurality of suffixes for the entity of interest;

create a plurality of individuals, wherein each of the plurality of individuals comprises a sequence of prefix words and a sequence of suffix words;

apply each of the plurality of individuals to each of the plurality of training documents to generate a fitness score associated with each of the plurality of individuals; and

convert an individual having the highest fitness score into a new rule.

6. The system of claim 5 , comprising code configured to direct the processor to annotate each token of each of the plurality of training documents with a parts-of-speech tag.

7. The system of claim 5 , wherein the storage system comprises code configured to direct the processor to:

evolve the plurality of individuals for a preselected number of generations; and

convert an individual rule having the highest fitness score into a rule.

8. The system of claim 7 , wherein the memory comprises code configured to direct the processor to:

remove suffixes associated with the rule from the plurality of suffixes;

remove prefixes associated with the rule from the plurality of prefixes;

generate a new plurality of individuals;

apply each of the new plurality of individuals to each of the plurality of documents to generate a fitness score associated with each of the new plurality of individuals;

evolve the new plurality of individuals for a preselected number of generations; and

convert an individual in the new plurality of individuals having the highest fitness score into a new rule.

9. The system of claim 5 , wherein evolving the plurality of individuals comprises:

calculating a fitness score for each individual in a population;

identifying a subset of individuals having a fitness score above a threshold;

discarding all of the individuals having fitness scores less than the threshold; and

recombining the individuals to form a new generation of individuals.

10. The system of claim 9 , wherein the threshold is selected to eliminate a portion of individuals from the population.

11. The system of claim 9 , wherein recombining the individuals comprises randomly selecting prefixes and suffixes from at least two individuals to create a new individual.

12. The system of claim 1 , comprising a network interface, wherein the storage system comprises code configured to direct the processor to obtain an unstructured document through the network interface.

13. The system of claim 1 , comprising a display, wherein the storage system comprises code configured to direct the processor to display an unstructured document, wherein the entities and extraction targets of the unstructured document are marked.

14. A method of extracting a target entity from a document, comprising:

performing an entity recognition on an unstructured document in a database to create found entities;

annotating the found entities in the unstructured document, wherein annotating comprises inserting tags around the found entities to generate marked entities;

applying a rule to each of the marked entities in the unstructured document to generate a confidence value for every entity, wherein the rule comprises a plurality of prefixes for a target entity and a plurality of suffixes for the target entity; and

selecting a marked entity with the highest confidence value as an extraction target.

15. The method of claim 14 , comprising generating the rule through the application of a genetic algorithm to a plurality of training documents.

16. The method of claim 15 , comprising annotating each token of the plurality of training documents and the unstructured document with a parts-of-speech tag.

17. The method of claim 15 , comprising generating additional rules by:

removing suffixes associated with the rule from a plurality of suffixes;

removing prefixes associated with the rule from a plurality of prefixes;

generating a new plurality of individuals;

applying each of the new plurality of individuals to each of the plurality of training documents to generate a fitness score associated with each of the new plurality of individuals;

evolve the new plurality of individuals for a preselected number of generations; and

convert an individual in the new plurality of individuals having the highest fitness score into a new rule.

18. A non-transitory, computer readable medium comprising code configured to direct a processor to:

load an unstructured document into a memory;

perform an entity recognition on the unstructured document;

annotate entities recognized in the unstructured document to generate marked entities;

apply a rule to the marked entities to generate a confidence value for every marked entity, wherein the rule comprises a plurality of prefixes for a target entity and a plurality of suffixes for the target entity; and

select the marked entity with a highest confidence value as an extraction target.

19. The non-transitory computer readable medium of claim 18 , comprising code configured to direct the processor to:

obtain a plurality of training documents, wherein an entity of interest is marked in each of the plurality of training documents;

generate a bag of words of prefixes for the entity of interest;

generate a bag of words of suffixes for the entity of interest;

create a plurality of individuals, wherein each of the plurality of individuals comprises a plurality of prefix words selected the bag of words of prefixes and a plurality of suffix words selected from the bag of words of suffixes;

apply each of the plurality of individuals to each of the plurality of training documents to generate a fitness score associated with each of the plurality of individuals; and

convert an individual having the highest fitness score into a new rule.

20. The non-transitory computer readable medium of claim 19 , comprising code configured to direct the processor to:

remove suffixes associated with the rule from the bag of words of prefixes;

remove prefixes associated with the rule from the bag of words of suffixes;

generate a new plurality of individuals by selecting a plurality of prefix words from the bag of words of prefixes and selecting a plurality of suffix words from the bag of words of suffixes;

apply each of the new plurality of individuals to each of the plurality of documents to generate a fitness score for each of the new plurality of individuals;

evolve the new plurality of individuals for a preselected number of generations; and

convert an individual in the new plurality of individuals having the highest fitness score into a new rule.