Intelligent extraction of information from a document

1. A computer-implemented method comprising:

receiving, by a first module of a computing system, input data representing an image of a document;

receiving, by the first module of the computing system, context data for the document, wherein the context data comprises parameters that are descriptive of the document in the image represented by the input data;

processing, by the first module of the computing system, the input data and the context data to determine a complexity value that characterizes a level of complexity in identifying information to be extracted from the document;

selecting, based on the complexity value and from a plurality of candidate machine-learning models, a particular machine-learning model to use in extracting information from the document; and

extracting information from the document using the particular selected machine-learning model, comprising converting a portion of the image of the document that shows typed or handwritten text into a digitized text string.

2. The method of claim 1 , wherein processing the input data and the context data comprises:

determining a quantity of labels, each label in the quantity of labels corresponding to a distinct portion of information in the document; and

identifying one or more reference templates that each correspond to the document based on the determined quantity of labels.

3. The method of claim 2 , wherein determining the complexity value comprises:

determining that the quantity of labels exceeds a threshold quantity;

identifying a particular reference template in response to determining that the quantity of labels exceeds the threshold quantity; and

determining the complexity value based on the particular reference template.

4. The method of claim 3 , wherein:

the particular reference template is a complex template comprising a plurality data fields,

each data field of the plurality of data fields corresponds to a respective label in the quantity of labels, and

each data field comprises a term corresponding to one or more terms in the extracted information that is converted into the digitized text string.

5. The method of claim 3 , wherein selecting the particular machine-learning model comprises:

determining, based on the complexity value, that the level of complexity in identifying information to be extracted from the document exceeds a threshold level; and

selecting the particular machine-learning model in response to determining that the level of complexity exceeds the threshold level.

6. The method of claim 5 , wherein selecting the particular machine-learning model comprises:

selecting a deep-learning model when (i) the quantity of labels exceeds the threshold quantity and (ii) the complexity value indicates the level of complexity in identifying information to be extracted from the document exceeds the threshold level.

7. The method of claim 5 , wherein selecting the particular machine-learning model comprises:

selecting a machine-learning classifier when (i) the quantity of labels is below the threshold quantity and (ii) the complexity value indicates the level of complexity in identifying information to be extracted from the document is below the threshold level.

8. The method of claim 5 , wherein:

selecting the particular machine-learning model comprises selecting the particular machine-learning model from a machine-learning engine,

the machine-learning engine includes at least two predictive models corresponding to the plurality of candidate machine-learning models, and

the at least two predictive models comprises: a machine-learning model that is operable to extract information relating to a region of interest in the image of the document; and a deep-learning model operable to read discrete terms in the region of interest or read discrete characters of the discrete terms in the region of interest.

9. The method of claim 8 , wherein converting the portion of the image of the document into the digitized text string comprises:

using the deep-learning model to convert the extracted information from the document in response to reading the discrete terms in the region of interest.

10. The method of claim 8 , wherein converting the portion of the image of the document into the digitized text string comprises:

using the deep-learning model to convert the extracted information from the document in response to reading the discrete characters of the discrete terms in the region of interest.

11. A system comprising:

one or more processing devices; and

one or more non-transitory machine-readable storage devices storing instructions that are executable by the one or more processing devices to cause performance of operations comprising:

receiving, by a first module of a computing system, input data representing an image of a document;

receiving, by the first module of the computing system, context data for the document, wherein the context data comprises parameters that are descriptive of the document in the image represented by the input data;

processing, by the first module of the computing system, the input data and the context data to determine a complexity value that characterizes a level of complexity in identifying information to be extracted from the document;

selecting, based on the complexity value and from a plurality of candidate machine-learning models, a particular machine-learning model to use in extracting information from the document; and

extracting information from the document using the particular selected machine-learning model, comprising converting a portion of the image of the document that shows typed or handwritten text into a digitized text string.

12. The system of claim 11 , wherein processing the input data and the context data comprises:

determining a quantity of labels, each label in the quantity of labels corresponding to a distinct portion of information in the document; and

identifying one or more reference templates that each correspond to the document based on the determined quantity of labels.

13. The system of claim 12 , wherein determining the complexity value comprises:

determining that the quantity of labels exceeds a threshold quantity;

identifying a particular reference template in response to determining that the quantity of labels exceeds the threshold quantity; and

determining the complexity value based on the particular reference template.

14. The system of claim 13 , wherein:

the particular reference template is a complex template comprising a plurality data fields,

each data field of the plurality of data fields corresponds to a respective label in the quantity of labels, and

each data field comprises a term corresponding to one or more terms in the extracted information that is converted into the digitized text string.

15. The system of claim 13 , wherein selecting the particular machine-learning model comprises:

determining, based on the complexity value, that the level of complexity in identifying information to be extracted from the document exceeds a threshold level; and

selecting the particular machine-learning model in response to determining that the level of complexity exceeds the threshold level.

16. The system of claim 15 , wherein selecting the particular machine-learning model comprises:

selecting a deep-learning model when (i) the quantity of labels exceeds the threshold quantity and (ii) the complexity value indicates the level of complexity in identifying information to be extracted from the document exceeds the threshold level.

17. The system of claim 15 , wherein selecting the particular machine-learning model comprises:

selecting a machine-learning classifier when (i) the quantity of labels is below the threshold quantity and (ii) the complexity value indicates the level of complexity in identifying information to be extracted from the document is below the threshold level.

18. The system of claim 15 , wherein:

selecting the particular machine-learning model comprises selecting the particular machine-learning model from a machine-learning engine,

the machine-learning engine includes at least two predictive models corresponding to the plurality of candidate machine-learning models, and

the at least two predictive models comprises: a machine-learning model that is operable to extract data for a region of interest in the image of the document; and a deep-learning model operable to read discrete terms included in the region of interest or read discrete characters of the discrete terms in the region of interest.

19. The system of claim 18 , wherein converting the portion of the image of the document into the digitized text string comprises:

using the deep-learning model to convert the extracted information from the document in response to reading the discrete terms in the region of interest; and

using the deep-learning model to convert the extracted information from the document in response to reading the discrete characters of the discrete terms in the region of interest.

20. One or more non-transitory machine-readable storage devices storing instructions that are executable by one or more processing devices to cause performance of operations comprising:

receiving, by a first module of a computing system, input data representing an image of a document;

receiving, by the first module of the computing system, context data for the document, wherein the context data comprises parameters that are descriptive of the document in the image represented by the input data;

processing, by the first module of the computing system, the input data and the context data to determine a complexity value that characterizes a level of complexity in identifying information to be extracted from the document;

selecting, based on the complexity value and from a plurality of candidate machine-learning models, a particular machine-learning model to use in extracting information from the document; and

extracting information from the document using the particular selected machine-learning model, comprising converting a portion of the image of the document that shows typed or handwritten text into a digitized text string.