Table row identification using machine learning

1. A method comprising:

detecting a table body in a document by processing the document using a machine learning (ML)-based table body model;

predicting an initial table row index for one or more words among a plurality of words obtained in the document, wherein the one or more words are determined to be within the table body;

determining a table row index for the one or more words using an ML-based table row model that is trained based on the predicted initial table row index for the one or more words, wherein the table row index is a real number; and

applying a rounding function to the table row index to determine a table row for the one or more words determined to be within the table body based on the table row index.

2. The method of claim 1 , wherein the table row model is a graph neural network (GNN) model.

3. The method of claim 1 , wherein the table row model is further trained using an error function that minimizes an error associated with determining a row transition based on the initial table row index for the one or more words determined to be within the table body.

4. The method of claim 1 , further comprising detecting a table header in the document by processing the document using an ML-based table header model.

5. The method of claim 4 , wherein the table row model is further trained based on determining whether one or more words among the obtained words are within the table header.

6. The method of claim 4 , wherein the table body model or the table header model is an ML-based computer vision model that is trained using supervised learning.

7. The method of claim 1 ,

wherein the document is selected among a corpus of documents, and

wherein the table row model is trained based on the initial table row index for the one or more words for the documents in the corpus.

8. The method of claim 1 ,

wherein the document is selected among a corpus of documents, each document in the corpus being annotated, and

wherein the plurality of words is obtained by processing the document using optical character recognition and an ML-based word model trained based on the annotated documents.

9. The method of claim 1 , wherein the plurality of words is obtained using a document-type-specific information extraction tool.

10. A system comprising:

at least one processing device including a processor coupled to a memory;

the at least one processing device being configured to implement the following steps:

detecting a table body in a document by processing the document using a machine learning (ML)-based table body model;

predicting an initial table row index for one or more words among a plurality of words obtained in the document, wherein the one or more words are determined to be within the table body;

determining a table row index for the one or more words using an ML-based table row model that is trained based on the predicted initial table row index for the one or more words, wherein the table row index is a real number; and

applying a rounding function to the table row index to determine a table row for the one or more words determined to be within the table body based on the table row index.

11. The system of claim 10 , wherein the table row model is further trained using an error function that minimizes an error associated with determining a row transition based on the initial table row index for the one or more words determined to be within the table body.

12. The system of claim 10 , wherein the processing device is further configured to implement detecting a table header in the document by processing the document using an ML-based table header model.

13. The system of claim 12 , wherein the table row model is further trained based on determining whether one or more words among the obtained words are within the table header.

14. The system of claim 12 , wherein the table body model or the table header model is an ML-based computer vision model that is trained using supervised learning.

15. The system of claim 10 ,

wherein the document is selected among a corpus of documents, each document in the corpus being annotated, and

wherein the plurality of words is obtained by processing the document using optical character recognition and an ML-based word model trained based on the annotated documents.

16. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:

detecting a table body in a document by processing the document using a machine learning (ML)-based table body model;

predicting an initial table row index for one or more words among a plurality of words obtained in the document, wherein the one or more words are determined to be within the table body;

determining a table row index for the one or more words using an ML-based table row model that is trained based on the predicted initial table row index for the one or more words, wherein the table row index is a real number; and

applying a rounding function to the table row index to determine a table row for the one or more words determined to be within the table body based on the table row index.

17. The storage medium of claim 16 , wherein the document is unstructured.

18. The storage medium of claim 16 , wherein the document is a purchase order.

19. The storage medium of claim 16 , wherein the document is an inventory manifest.

20. The storage medium of claim 16 , wherein the table body model is an ML-based deep neural network (DNN) model.