System for deep learning using knowledge graphs

1. A method comprising:

receiving data from a computing device requesting approval of a loan application;

accessing time-series data associated with a user associated with the loan application from a knowledge graph, the time-series data including real description format (RDF) triples that identify account balances over a period of time;

building a multi-dimensional feature vector based on the RDF triples of the accessed time-series data; wherein building the multi-dimensional feature vector includes converting the RDF triples into a vector format, wherein a component of the multi-dimensional feature vector includes an account balance of the user of the RDF triples that identify account balances over a period of time;

inputting the multi-dimensional feature vector into a machine learning model, the machine learning model configured as a recurrent neural network having a plurality of output nodes indicating a respective plurality of reasons for rejecting the loan application;

receiving an output from the machine learning model in response to the inputting, the output indicating a level of approval for the user with respect to the loan application; and

transmitting a response to the request based on the level of approval.

2. The method of claim 1 , further comprising:

accessing at least one value from the plurality of output nodes after inputting the feature vector into the machine learning model;

transmitting the at least one value from the plurality of output nodes for verification;

receiving a response to the verification, the response indicating whether the at least one value is valid for the inputted feature vector; and

wherein the transmitted response to the requested approval is further based on the response to the verification.

3. The method of claim 2 , further comprising

training the machine learning model based on the response to the verification.

4. The method of claim 1 , further comprising:

requesting a reason for rejection of the loan application when the level of approval is below a threshold;

building a labeled feature vector based on the time-series data and a response to the request for the reason for rejection; and

training a second machine learning model based on the labeled feature vector.

5. The method of claim 1 , wherein coefficients of nodes in the machine learning model are based on:

feature vectors based on performance of loans of previously accepted loan applications; and

feature vectors based on inferred performance of previously denied loan applications.

6. The method of claim 1 , wherein data in the knowledge graph is based on data stored in a relational database system translated into a triplestore database.

7. A non-transitory computer readable medium comprising instructions, which when executed by at least one processor, configure the at least one processor to perform operations comprising:

receiving data from a computing device requesting approval of a loan application;

accessing time-series data associated with a user associated with the loan application from a knowledge graph, the time-series data including real description format (RDF) triples that identify account balances over a period of time;

building a multi-dimensional feature vector based on the RDF triples of the accessed time-series data, wherein building the multi-dimensional feature vector includes converting the RDF triples into a vector format, wherein a component of the multi-dimensional feature vector includes an account balance of the user of the RDF triples that identify account balances over a period of time;

inputting the multi-dimensional feature vector into a machine learning model, the machine learning model configured as a recurrent neural network having a plurality of output nodes indicating a respective plurality of reasons for rejecting the loan application;

receiving an output from the machine learning model in response to the inputting, the output indicating a level of approval for the user with respect to the loan application; and

transmitting a response to the request based on the level of approval.

8. The non-transitory computer readable medium of claim 7 , the operations further comprising:

accessing at least one value from the plurality of output nodes after inputting the feature vector into the machine learning model;

transmitting the at least one value from the plurality of output nodes for verification;

receiving a response to the verification, the response indicating whether the at least one value is valid for the inputted feature vector; and

wherein the transmitted response to the requested approval is further based on the response to the verification.

9. The non-transitory computer readable medium of claim 8 , the operations further comprising

training the machine learning model based on the response to the verification.

10. The non-transitory computer readable medium of claim 7 , the operations further comprising:

requesting a reason for rejection of the loan application when the level of approval is below a threshold;

building a labeled feature vector based on the time-series data and a response to the request for the reason for rejection; and

training a second machine learning model based on the labeled feature vector.

11. The non-transitory computer readable medium of claim 7 , wherein coefficients of nodes in the machine learning model are based on:

feature vectors based on performance of loans of previously accepted loan applications; and

feature vectors based on inferred performance of previously denied loan applications.

12. The non-transitory computer readable medium of claim 7 , wherein data in the knowledge graph is based on data stored in a relational database system translated into a triplestore database.

13. A system comprising:

at least one processor; and

a storage device comprising instructions, which when executed by the at least one processor, configure at the at least one processor to:

receive data from a computing device requesting approval of a loan application;

access time-series data associated with a user associated with the loan application from a knowledge graph, the time-series data including real description format (RDF) triples that identify account balances over a period of time;

build a multi-dimensional feature vector based on the RDF triples of the accessed time-series data, wherein building the multi-dimensional feature vector includes converting the RDF triples into a vector format, wherein a component of the multi-dimensional feature vector includes an account balance of the user of the RDF triples that identify account balances over a period of time;

input the multi-dimensional feature vector into a machine learning model, the machine learning model configured as a recurrent neural network having a plurality of output nodes indicating a respective plurality of reasons for rejecting the loan application;

receive an output from the machine learning model in response to the inputting, the output indicating a level of approval for the user with respect to the loan application; and

transmit a response to the request based on the level of approval.

14. The system of claim 13 , wherein the at least one processor is further configured, when executing the instructions, to:

access at least one value from the plurality of output nodes after inputting the feature vector into the machine learning model;

transmit the at least one value from the plurality of output nodes for verification;

receive a response to the verification, the response indicating whether the at least one value is valid for the inputted feature vector; and

wherein the transmitted response to the requested approval is further based on the response to the verification.

15. The system of claim 14 , wherein the at least one processor is further configured, when executing the instructions, to:

train the machine learning model based on the response to the verification.

16. The system of claim 13 , wherein the at least one processor is further configured, when executing the instructions, to:

request a reason for rejection of the loan application when the level of approval is below a threshold;

build a labeled feature vector based on the time-series data and a response to the request for the reason for rejection; and

training a second machine learning model based on the labeled feature vector.

17. The system of claim 13 , wherein coefficients of nodes in the machine learning model are based on:

feature vectors based on performance of loans of previously accepted loan applications; and

feature vectors based on inferred performance of previously denied loan applications.

18. The system of claim 13 , wherein data in the knowledge graph is based on data stored in a relational database system translated into a triplestore database.