Automated system and method for electronic document type identification and validation

1 . A method of operating a document validation system, the method comprising:

extracting words and phrases from content associated with a submitted, electronic document image;

attempting to classify a document type for the electronic document image using a first machine learning model of one or more machine learning models, wherein the document type includes one of an identification and a financial instrument;

attempting to verify the electronic document image has been classified as the one of the identification and the financial instrument, and if not verified, attempting to classify the document type using a machine learning model of the one or more machine learning models that is different from the first machine learning model to verify the document type;

identifying key value pairs from the extracted words and phrases using word embedding vectors (WEVs) and cosine similarity calculations;

determining the document validity using the key value pairs with one or more rules-based models;

returning document validity and verified document type in response to determining the document validity and verifying the type of the classified document; and

returning an error result in response to failing to determine the document validity or failure to verify the type of the classified document.

2 . The method of claim 1 further comprising:

submitting a document image through a portal;

passing non-rejected document images to an application; and

requesting validation of the document from the document validation system through an application programming interface.

3 . The method of claim 2 further comprising:

rejecting document images with inapplicable content.

4 . The method of claim 1 , wherein the attempting to classify operation further comprises:

running the machine learning models at least in part on an integrated circuit optimized for running machine learning models.

5 . The method of claim 1 , wherein the identifying operation further comprises:

providing a list of key phrases, each key phrase comprising one or more words;

generating word embedding vectors for the key phrases;

generating word embedding vectors for the extracted words and phrases;

calculating the cosine distance between the word embedding vectors of the key phrases and the word embedding vectors of the extracted words and phrases; and

identifying the extracted words and phrases of which the minimum cosine distance is less than a prescribed threshold with respect to the key phrases as key value pairs.

6 . The method of claim 5 , wherein:

the word embedding vectors for the key phrases are generated using a pre-trained model; and

the word embedding vectors for the extracted words and phrases are generated using the pre-trained model.

7 . The method of claim 1 , wherein the determining operation further comprises:

extracting dates from between key value pairs;

determining the recency of the document; and

validating the document relative to the recency and a submission date.

8 . The method of claim 1 , wherein:

the types of documents to be classified and validated comprise at least one of the list consisting of: pay stubs, bank statements, driver's licenses, passports, leases, mortgages, and contracts.

9 . The method of claim 1 , wherein:

user metadata is used with the rules-based models to determine document validity.

10 . The method of claim 1 , wherein:

a document image is submitted from a computing device through a portal;

the portal sends the document image to an application; and

the application submits the document image to the document validation system through an application programming interface.

11 . The method of claim 10 , wherein:

document validity and verified document type are returned through the application programming interface.

12 . The method of claim 1 , wherein:

the one or more machine learning models are applied in series until a document is successfully classified.

13 . The method of claim 1 , wherein:

the one or more machine learning models are applied in parallel.

14 . A device, comprising:

a processor;

a memory; and

a non-transitory storage medium comprising machine executable instructions executable by the processor, further comprising:

first instructions for extracting words and phrases from content from a submitted document image through an application programming interface,

second instructions for classifying a document type of the submitted document image using one or more machine learning models, wherein the document type comprises one of an identification including a government identification and a financial instrument including a bank statement, a first machine learning model of the one or more machine learning models is directed to classifying the submitted document image as the bank statement and a second machine learning model of the one or more machine learning models is directed to classifying the submitted document image as the government identification,

third instructions for identifying key value pairs from the extracted words and phrases using word embedding vectors (WEVs) and cosine similarity calculations,

fourth instructions for determining the document validity using the key value pairs with one or more rules-based models,

fifth instructions for verifying the type of the classified document, and

sixth instructions for returning document validity and verified document type.

15 . The device of claim 14 , wherein the processor is optimized for running machine learning models.

16 . The device of claim 14 , wherein the third instructions further comprise instructions for:

providing a list of key phrases, each key phrase comprising one or more words;

generating word embedding vectors for the key phrases;

generating word embedding vectors for the extracted words and phrases;

calculating the cosine distance between the word embedding vectors of the key phrases and the word embedding vectors of the extracted words and phrases; and

identifying the extracted words and phrases of which the minimum cosine distance is less than a prescribed threshold with respect to the key phrases as key value pairs.

17 . The device of claim 16 , wherein:

the word embedding vectors for the key phrases are generated using a pre-trained model; and

the word embedding vectors for the extracted words and phrases are generated using the pre-trained model.

18 . The device of claim 14 , wherein:

the document image is passed to the device through an application programming interface; and

the document validity and verified document type are returned through the application programming interface.

19 . A non-transitory storage medium comprising machine executable instructions, further comprising:

first instructions for extracting words and phrases from content from a submitted document image through an application programming interface,

second instructions for classifying a document type of the submitted document image using one or more machine learning models, wherein each of the one or more machine learning models is directed to classifying a different document type and the document type includes an identification and a financial instrument,

third instructions for identifying key value pairs from the extracted words and phrases using word embedding vectors (WEVs) and cosine similarity calculations,

fourth instructions for determining the document validity using the key value pairs with one or more rules-based models,

fifth instructions for verifying the type of the classified document, and

sixth instructions for returning document validity and verified document type.

20 . The non-transitory storage medium of claim 19 , wherein the third instructions further comprise instructions for:

providing a list of key phrases, each key phrase comprising one or more words;

generating word embedding vectors for the key phrases;

generating word embedding vectors for the extracted words and phrases;

calculating the cosine distance between the word embedding vectors of the key phrases and the word embedding vectors of the extracted words and phrases; and

identifying the extracted words and phrases of which the minimum cosine distance is less than a prescribed threshold with respect to the key phrases as key value pairs.