Methods and systems for artificial intelligence-assisted document annotation
Methods and systems for artificial intelligence (AI)-assisted document annotation and training of machine learning-based models for document data extraction are described. The methods and systems described herein take advantage of a continuous machine learning approach to create document processing pipelines that provide accurate and efficient data extraction from documents that include structured text, semi-structured text, unstructured text, or any combination thereof.
1. A computer-implemented method for annotating an electronic document comprising:
displaying, within a first region of a graphical user interface, an electronic document, or a page therefrom;
displaying, within a second region of the graphical user interface, a list of suggested labels that may be applicable to categories of text within the electronic document;
receiving a first input from a user indicating a first selection of text within the first region of the graphical user interface;
receiving a second input from the user to assign a first label from the list of suggested labels to the selected text;
displaying, within the first region of the graphical user interface, a graphic element comprising the assigned first label and the first selection of text, wherein the graphic element is adjacent to, or overlaid on, a location of the first selection of text;
storing the assigned first label, the first selection of text, and the location of the first selection of text for one or more instances of the first selection of text within the electronic document as an annotated electronic document;
receiving a third input from the user indicating a second selection of text within the first region of the graphical user interface;
receiving a fourth input from the user to assign a second label from the list of suggested labels to the second selection of text;
displaying, within the first region of the graphical user interface, a graphic element comprising the assigned second label and the second selection of text;
storing the assigned second label, the second selection of text, and the location of the second selection of text for one or more instances of the second selection of text within the annotated electronic document; and
using the annotated electronic document to train a first machine learning model to extract text corresponding to the first label and to train a second machine learning model to extract text corresponding to the second label, wherein the first and second machine learning models are stored in a repository of user-selectable machine learning models for performing data extraction from electronic documents.
2. The computer-implemented method of claim 1 , further comprising displaying, within the first region of the graphical user interface, suggested selections of text that may correspond to the suggested labels.
3. The computer-implemented method of claim 1 , further comprising repeating the steps of receiving user input for selections of text and receiving user input to assign labels for one or more additional selections of text and assigned labels.
4. The computer-implemented method of claim 1 , further comprising receiving a third input from the user to assign a custom label to the selection of text.
5. The computer-implemented method of claim 1 , wherein the selected text comprises a word, a phrase, a sentence, a paragraph, a section, or a table.
6. The computer-implemented method of claim 1 , wherein the list of suggested labels comprises a list of text categories that includes name, date, execution date, effective date, expiration date, delivery date, due date, date of sale, order date, invoice date, issuance data, address, address line 1 , street address, quantity, amount, cost, cost of goods sold, signature, or any combination thereof.
7. The computer-implemented method of claim 1 , further comprising displaying, within a third region of the graphical user interface, a list of selected text grouped according to assigned label.
8. The computer-implemented method of claim 1 , further comprising repeating the method for one or more additional electronic documents and storing the one or more additional annotated electronic documents.
9. The computer-implemented method of claim 1 , further comprising:
using the trained first machine learning model and the trained second machine learning model to predict selections of text corresponding to the first and second labels from one or more non-annotated validation electronic documents;
sequentially displaying each of the one or more validation electronic documents, or pages therefrom, in the first region of the graphical user interface, wherein the predictions of text corresponding to the first and second labels are graphically highlighted;
sequentially receiving feedback from the user on accuracy of the predicted selections of text corresponding to the first and second labels in each of the one or more validation electronic documents; and
approving or correcting each of the one or more validation electronic documents according to the feedback from the user.
10. The computer-implemented method of claim 9 , further comprising retraining the first or second machine learning model using the one or more approved or corrected validation electronic documents.
11. A system comprising:
one or more processors;
a memory;
an electronic display device; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for:
displaying, within a first region of a graphical user interface, an electronic document, or a page therefrom;
displaying, within a second region of the graphical user interface, a list of suggested labels that may be applicable to categories of text within the electronic document;
receiving a first input from a user indicating a first selection of text within the first region of the graphical user interface;
receiving a second input from the user to assign a first label from the list of suggested labels to the selected text;
displaying, within the first region of the graphical user interface, a graphic element comprising the assigned first label and the first selection of text, wherein the graphic element is adjacent to, or overlaid on, a location of the first selection of text;
storing the assigned first label, the first selection of text, and the location of the first selection of text for one or more instances of the first selection of text within the electronic document as an annotated electronic document;
receiving a third input from the user indicating a second selection of text within the first region of the graphical user interface;
receiving a fourth input from the user to assign a second label from the list of suggested labels to the second selection of text;
displaying, within the first region of the graphical user interface, a graphic element comprising the assigned second label and the second selection of text;
storing the assigned second label, the second selection of text, and the location of the second selection of text for one or more instances of the second selection of text within the annotated electronic document; and
using the annotated electronic document to train a first machine learning model to extract text corresponding to the first label and to train a second machine learning model to extract text corresponding to the second label, wherein the first and second machine learning models are stored in a repository of user-selectable machine learning models for performing data extraction from electronic documents.
12. The system of claim 11 , wherein the instructions further comprise displaying within the first region of the graphical user interface, suggested selections of text that may correspond to the suggested labels.
13. The system of claim 11 , wherein the instructions further comprise repeating the steps of receiving user input for selections of text and receiving user input to assign labels for one or more additional selections of text and assigned labels.
14. The system of claim 11 , wherein the instructions further comprise receiving a third input from the user to assign a custom label to the selection of text.
15. The system of claim 11 , wherein the selected text comprises a word, a phrase, a sentence, a paragraph, a section, or a table.
16. The system of claim 11 , wherein the list of suggested labels comprises a list of text categories that includes name, date, execution date, effective date, expiration date, delivery date, due date, date of sale, order date, invoice date, issuance data, address, address line 1 , street address, quantity, amount, cost, cost of goods sold, signature, or any combination thereof.
17. The system of claim 11 , wherein the instructions further comprise displaying, within a third region of the graphical user interface on the electronic display, a list of selected text grouped according to assigned label.
18. The system of claim 11 , wherein the instructions further comprise repeating the displaying, receiving, and storing steps for one or more additional electronic documents and storing one or more additional annotated electronic documents.
19. The system of claim 11 , wherein the instructions further comprise:
using the trained first machine learning model and the trained second machine learning model to predict selections of text corresponding to the first and second labels from one or more non-annotated validation electronic documents;
sequentially displaying each of the one or more validation electronic documents, or pages therefrom, in the first region of the graphical user interface, wherein the predicted selections of text corresponding to the first and second labels are graphically highlighted;
sequentially receiving feedback from the user on accuracy of the predicted selections of text corresponding to the first and second labels in each of the one or more validation electronic documents; and
approving or correcting the one or more validation electronic documents according to the feedback from the user.
20. The system of claim 19 , wherein the instructions further comprise retraining the first or second machine learning model using the one or more approved or corrected validation electronic documents.
21. A non-transitory computer-readable storage medium storing one or more programs, the one or more programs comprising instructions, the instructions when executed by one or more processors of a computing platform, cause the computing platform to:
display, within a first region of a graphical user interface, an electronic document, or a page therefrom;
display, within a second region of the graphical user interface, a list of suggested labels that may be applicable to categories of text within the electronic document;
receive a first input from a user indicating a first selection of text within the first region of the graphical user interface;
receive a second input from the user to assign a first label from the list of suggested labels to the selected text;
display, within the first region of the graphical user interface, a graphic element comprising the assigned first label and the first selection of text, wherein the graphic element is adjacent to, or overlaid on, a location of the first selection of text;
store the assigned first label, the first selection of text, and the location of the first selection of text for one or more instances of the first selection of text within the electronic document as an annotated electronic document;
receiving a third input from the user indicating a second selection of text within the first region of the graphical user interface;
receiving a fourth input from the user to assign a second label from the list of suggested labels to the second selection of text;
displaying, within the first region of the graphical user interface, a graphic element comprising the assigned second label and the second selection of text;
storing the assigned second label, the second selection of text, and the location of the second selection of text for one or more instances of the second selection of text within the annotated electronic document; and
using the annotated electronic document to train a first machine learning model to extract text corresponding to the first label and to train a second machine learning model to extract text corresponding to the second label, wherein the first and second machine learning models are stored in a repository of user-selectable machine learning models for performing data extraction from electronic documents.
22. The non-transitory computer-readable storage medium of claim 21 , wherein the instructions further comprise displaying within the first region of the graphical user interface, suggested selections of text that may correspond to the suggested labels.
23. The non-transitory computer-readable storage medium of claim 21 , wherein the instructions further comprise repeating the steps of receiving user input for selections of text and receiving user input to assign labels for one or more additional selections of text and assigned labels.
24. The non-transitory computer-readable storage medium of claim 21 , wherein the instructions further comprise receiving a third input form the user to assign a custom label to the selection of text.
25. The non-transitory computer-readable storage medium of claim 21 , wherein the instructions further cause the computing platform to:
use the trained first machine learning model and the trained second machine learning model to predict selections of text corresponding to the first and second labels from one or more non-annotated validation electronic documents;
sequentially display each of the one or more validation electronic documents, or pages therefrom, in the first region of the graphical user interface, wherein the predicted selections of text corresponding to the first and second labels are graphically highlighted;
sequentially receive feedback from the user on accuracy of the predicted selections of text corresponding to the first and second labels in each of the one or more validation electronic documents; and
approve or correct the one or more validation electronic documents according to the feedback from the user.