Predicting outlier data from network of electronic data
A computer-implemented method, computer program product, and/or computing system for determining unexpected data includes: running for a given claim a machine learning term prediction model with first and second term types to predict third type terms in a specific domain that uses a machine learning term embedding model that has learned the correlation between term types in the specific domain; determining cohesiveness between the predicted third type terms and the actual third type terms from the given claim; determining a value score between the predicted third type terms and the actual third type; and running a machine learning propensity model using the value score determined for the given claim and context data for the given claim to determine if the given claim is unexpected.
1. A computer implemented method, the method comprising:
building and training a machine learning term embedding model for a specific domain to learn the correlation between at least a first term type, a second term type and a third claim type using historical data from the specific domain;
building and training a machine learning term prediction model using the term embedding model and historical data to predict terms used in claims in the specific domain;
extracting first type terms, second type terms, and third type terms for a given claim in the specific domain;
running the term prediction model with first term types and second term types extracted from the given claim in the specific domain to obtain predicted third type terms for the given claim;
determining cohesiveness between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
determining a value score between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
building and training a machine learning propensity model using the value score and context data from historical corpus of claims to predict the likelihood that the third type terms are unexpected; and
running the propensity model with the value score determined for the given claim and context data for the given claim to determine if the given claim is unexpected.
2. The method of claim 1 , further comprising extracting at least one of an extraction group consisting of lexicons, key words, and combinations thereof from unstructured and structured data from documents in the specific domain to build and train the term embedding model.
3. The method of claim 2 , further comprising using Natural Language Processing to extract at least one of the extraction group.
4. The method of claim 1 , wherein determining the cohesiveness between the predicted third type terms extracted from the given claim comprises using at least one of a measurement group consisting of clustering techniques, distance in a term vector space, and combinations thereof.
5. The method of claim 1 , wherein determining the cohesiveness between the predicted third type terms extracted from the given claim comprises predicting other third type terms from a subset of third type terms.
6. The method of claim 1 , further comprising determining a cohesiveness score between the predicted third type terms and the third type terms extracted from the given claim.
7. The method of claim 6 , wherein determining value score between the predicted third type terms and the third type terms extracted from the given claim comprises using the cohesiveness score and the output from the term prediction model.
8. The method of claim 1 , wherein the specific domain is vehicle accidents, the first term type are terms descriptive of the vehicle accident, the second term type are terms descriptive of the vehicle damage, and the third term type are terms descriptive of the injury.
9. The method of claim 1 , further comprising identifying unexpected portions of the given claim.
10. The method of claim 1 , further comprising scoring the given claim.
11. A computer programing product comprising a computer readable storage medium storing instructions that, when executed by at least one hardware processor, configure the at least one hardware processor to:
build and train a machine learning term embedding model for a specific domain to learn the correlation between at least a first term type, a second term type and a third claim type using historical data from the specific domain;
build and train a machine learning term prediction model using the term embedding model and historical data to predict terms used in claims in the specific domain;
extract first type terms, second type terms, and third type terms for a given claim in the specific domain;
run the term prediction model with first term types and second term types extracted from the given claim in the specific domain to obtain predicted third type terms for the given claim;
determine cohesiveness between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
determine value score between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
build and train a machine learning propensity model using the value score and context data from historical corpus of claims to predict the likelihood that the third type terms are unexpected; and
run propensity model with the value score determined for the given claim and context data for the given claim to determine if the given claim is unexpected.
12. The computer program product according to claim 11 , further comprising instructions that, when executed by the at least one hardware processor, configure the at least one hardware processor to extract at least one of an extraction group consisting of lexicons, key words, and combinations thereof from unstructured and structured data from documents in the specific domain to build and train the term embedding model.
13. The computer program product according to claim 12 , further comprising instructions that, when executed by the at least one hardware processor, configure the at least one hardware processor to use Natural Language Processing to extract at least one of the extraction group.
14. The computer program product according to claim 11 , wherein instructions to determine the cohesiveness between the predicted third type terms extracted from the given claim comprises instructions that, when executed by the at least one hardware processor, configure the at least one hardware processor to use at least one of a measurement group consisting of clustering techniques, distance in a term vector space, and combinations thereof.
15. The computer program product according to claim 11 , further comprising instructions that, when executed by the at least one hardware processor, configure the at least one hardware processor to determine a cohesiveness score between the predicted third type terms and the third type terms extracted from the given claim.
16. The computer program product according to claim 15 , wherein instructions to determine a value score between the predicted third type terms and the third type terms extracted from the given claim comprises instructions that, when executed by the at least one hardware processor, configure the at least one hardware processor to use the cohesiveness score and the output from the term prediction model.
17. The computer program product according to claim 11 , wherein the specific domain is vehicle accidents, the first term type are terms descriptive of the vehicle accident, the second term type are terms descriptive of the vehicle damage, and the third term type are terms descriptive of the injury.
18. A system comprising:
a processor; and
an outlier prediction module wherein the outlier prediction module configures the processor to:
build and train a machine learning term embedding model for a specific domain to learn the correlation between at least a first term type, a second term type and a third claim type using historical data from the specific domain;
build and train a machine learning term prediction model using the term embedding model and historical data to predict terms used in claims in the specific domain;
extract first type terms, second type terms, and third type terms for a given claim in the specific domain;
run the term prediction model with first term types and second term types extracted from the given claim in the specific domain to obtain predicted third type terms for the given claim;
determine cohesiveness between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
determine value score between the predicted third type terms and the third type terms extracted from the given claim in the specific domain;
build and train a machine learning propensity model using the value score and context data from historical corpus of claims to predict the likelihood that the third type terms are unexpected; and
run propensity model with the value score determined for the given claim and context data for the given claim to determine if the given claim is unexpected.
19. The system of claim 18 , wherein the outlier prediction module further configures the processor to extract at least one of an extraction group consisting of lexicons, key words, and combinations thereof from unstructured and structured data from documents in the specific domain to build and train the term embedding model.
20. The system of claim 19 , wherein the specific domain is vehicle accidents, the first term type are terms descriptive of the vehicle accident, the second term type are terms descriptive of the vehicle damage, and the third term type are terms descriptive of the injury, and
wherein configuring the processor to determine the cohesiveness between the predicted third type terms extracted from the given claim further comprises the outlier prediction module further configuring the processor to: use at least one of a measurement group consisting of clustering techniques, distance in a term vector space, and combinations thereof; and determine a cohesiveness score between the predicted third type terms and the third type terms extracted from the given claim; and
wherein configuring the processor to determining value score between the predicted third type terms and the third type terms extracted from the given claim further comprises the outlier prediction module further configuring the processor to use the cohesiveness score and the output from the term prediction model.