Machine learning modification and natural language processing
A system, computer program product, and method are provided to automate a framework for knowledge graph based persistence of data, and to resolve temporal changes and uncertainties in the knowledge graph. Natural language understanding, together with one or more machine learning models (MLMs), is used to extract data from unstructured information, including entities and entity relationships. The extracted data is populated into a knowledge graph. As the KG is subject to change, the KG is used to create new and retrain existing machine learning models (MLMs). Weighting is applied to the populated data in the form of veracity value. Blockchain technology is applied to the populated data to ensure reliability of the data and to provide auditability to assess changes to the data.
1. A computer system comprising:
a processing unit operatively coupled to memory;
an artificial intelligence platform, in communication with the processing unit and memory;
a knowledge engine operatively coupled to the processing unit to train a machine learning model (MLM), the knowledge engine configured to:
select a first MLM from a natural language (NL) processing library of MLMs, aligned to a knowledge domain expressed in a first knowledge graph (KG);
receive NL input and query the input against the first KG, and extract one or more triplets from the first KG;
apply the selected MLM to a second KG different from the first KG, and extract one or more triplets from the second KG, wherein each triplet includes a subject, object, and a relationship;
for each extracted triplet:
obtain a blockchain (BC) identifier associated with each triplet; and
identify a triplet veracity value from a corresponding BC ledger;
detect a modification of the first KG from the extracted one or more triplets from the second KG, wherein the modification is selected from the group consisting of: content and structure, and combinations thereof; and
evaluate the detected modification, including employ the obtained BC identifier to assess veracity of the detected modification; and
dynamically augment the first MLM responsive to the received NL input.
2. The system of claim 1 , wherein the detected modification is content, and further comprising the knowledge engine to classify the detected modification, wherein the classification is selected from the group consisting of: synchronic and diachronic.
3. The system of claim 2 , wherein the detected modification is classified as conflicting data, and further comprising the knowledge engine to leverage the assessed veracity value of the first and second data, and limit modification of the first MLM subject to the assessed veracity value.
4. The system of claim 2 , further comprising the knowledge engine to employ the classification as a contribution factor with the modification evaluation.
5. The system of claim 1 , wherein the dynamic modification augmentation of the first MLM includes the MLM to create a new MLM.
6. A computer program product to process natural language (NL), the computer program product comprising a computer readable storage device having program code embodied therewith, the program code executable by a processing unit to:
select a first machine learning model (MLM) from a NL processing library of MLMs, aligned to a knowledge domain expressed in a first knowledge graph (KG);
receive NL input and query the input against the first KG, and extract one or more triplets from the first KG;
apply the selected MLM to a second KG different from the first KG, and extract one or more triplets from the second KG, wherein each triplet includes a subject, object, and a relationship, and for each extracted triplet:
obtain a blockchain (BC) identifier associated with each triplet; and
identify a triplet veracity value from a corresponding BC ledger;
detect a modification of the first KG from the extracted one or more triplets from the second KG, wherein the modification is selected from the group consisting of: content and structure, and combinations thereof;
evaluate the detected modification, including employ the obtained BC identifier to assess veracity of the detected modification; and
dynamically augment the first MLM responsive to the received NL input.
7. The computer program product of claim 6 , wherein the detected modification is content, and further comprising program code to:
classify the detected modification, wherein the classification is selected from the group consisting of: synchronic and diachronic.
8. The computer program product of claim 7 , further comprising program code to employ the classification as a contribution factor with the modification evaluation.
9. The computer program product of claim 7 , wherein the detected modification is classified as conflicting data, and further comprising program code to:
leverage the assessed veracity value of the first and second data, and limit modification of the first MLM subject to the assessed veracity value.
10. The computer program product of claim 6 , wherein the dynamic augmentation of the first MLM includes the MLM to create a new MLM.
11. A method for processing natural language (NL), comprising:
selecting a first machine learning model (MLM) from a NL processing library of MLMs, aligned to a knowledge domain expressed in a first knowledge graph (KG);
receiving NL input and query the input against the first KG, and extracting one or more triplets from the first KG;
applying the selected MLM to a second KG different from the first KG, and extracting one or more triplets from the second KG, wherein each triplet includes a subject, object, and a relationship, and for each extracted triplet:
obtaining a blockchain (BC) identifier associated with each triplet; and
identifying a triplet veracity value from a corresponding BC ledger;
detecting a modification of the first KG from the extracted one or more triplets from the second KG, wherein the modification is selected from the group consisting of: content and structure, and combinations thereof;
evaluating the detected modification, including employing the obtained BC identifier to assess veracity of the detected modification; and
dynamically augmenting the first MLM responsive to the received NL input.
12. The method of claim 11 , wherein the detected modification is content, and further comprising:
classifying the detected modification, wherein the classification is selected from the group consisting of: synchronic and diachronic.
13. The method of claim 12 , further comprising employing the classification as a contribution factor with the modification evaluation.
14. The method of claim 12 , wherein the detected modification is classified as conflicting data, and further comprising:
leveraging the assessed veracity value of the first and second data, and limiting modification of the first MLM subject to the assessed veracity value.
15. The method of claim 11 , wherein the dynamic augmentation of the first MLM includes the MLM creating a new MLM.