Contextual vector grids for document processing
Disclosed herein are system, method, and computer program product embodiments for analyzing contextual symbol information for document processing. In an embodiment, a language model system may generate a vector grid that incorporates contextual document information. The language model system may receive a document file and identify symbols of the document file to generate a symbol grid. The language model system may also identify position parameters corresponding to each of the symbols. The language model system may then analyze the symbols using an embedding function and neighboring symbols to determine contextual vector values corresponding to each of the symbols. The language model system may then generate a vector grid mapping the contextual vector values using the position parameters. The contextual information from the vector grid may provide increase document processing accuracy as well as faster processing convergence.
1. A computer implemented method, comprising:
receiving a document file;
generating a symbol grid including one or more symbols of the document file;
determining one or more position parameters corresponding to each of the one or more symbols;
analyzing the one or more symbols using an embedding function and one or more neighboring symbols to determine contextual vector values corresponding to each of the one or more symbols, wherein each of the contextual vector values represents respective semantic meanings for each of the one or more symbols based on the one or more neighboring symbols; and
generating a vector grid by mapping the contextual vector values with the one or more position parameters.
2. The computer implemented method of claim 1 , wherein the embedding function uses a neural network language model.
3. The computer implemented method of claim 1 , wherein the embedding function uses a Bidirectional Encoder Representations from Transformers (BERT) model.
4. The computer implemented method of claim 1 , wherein the document file is an image of a document and the vector grid provides two-dimensional layout information of the image of the document.
5. The computer implemented method of claim 1 , wherein the analyzing further comprises:
identifying a radius of neighboring symbols; and
selecting the one or more neighboring symbols within the radius.
6. The computer implemented method of claim 1 , wherein the symbol grid includes a first symbol and a second symbol having a common symbol value and wherein a first contextual vector value corresponding to the first symbol differs from a second contextual vector value corresponding to the second symbol.
7. The computer implemented method of claim 1 , further comprising:
performing a key value extraction from the document file using the vector grid.
8. A system, comprising:
a memory; and
at least one processor coupled to the memory and configured to:
receive a document file;
generate a symbol grid including one or more symbols of the document file;
determine one or more position parameters corresponding to each of the one or more symbols;
analyze the one or more symbols using an embedding function and one or more neighboring symbols to determine contextual vector values corresponding to each of the one or more symbols, wherein each of the contextual vector values represents respective semantic meanings for each of the one or more symbols based on the one or more neighboring symbols; and
generate a vector grid by mapping the contextual vector values with the one or more position parameters.
9. The system of claim 8 , wherein the embedding function uses a neural network language model.
10. The system of claim 8 , wherein the embedding function uses a Bidirectional Encoder Representations from Transformers (BERT) model.
11. The system of claim 8 , wherein the document file is an image of a document and the vector grid provides two-dimensional layout information of the image of the document.
12. The system of claim 8 , wherein to analyze the one or more symbols, the at least one processor is further configured to:
identify a radius of neighboring symbols; and
select the one or more neighboring symbols within the radius.
13. The system of claim 8 , wherein the symbol grid includes a first symbol and a second symbol having a common symbol value and wherein a first contextual vector value corresponding to the first symbol differs from a second contextual vector value corresponding to the second symbol.
14. The system of claim 8 , wherein the at least one processor is further configured to:
perform a key value extraction from the document file using the vector grid.
15. A non-transitory computer-readable device having instructions stored thereon that, when executed by at least one computing device, cause the at least one computing device to perform operations comprising:
receiving a document file;
generating a symbol grid including one or more symbols of the document file;
determining one or more position parameters corresponding to each of the one or more symbols;
analyzing the one or more symbols using an embedding function and one or more neighboring symbols to determine contextual vector values corresponding to each of the one or more symbols, wherein each of the contextual vector values represents respective semantic meanings for each of the one or more symbols based on the one or more neighboring symbols; and
generating a vector grid by mapping the contextual vector values with the one or more position parameters.
16. The non-transitory computer-readable device of claim 15 , wherein the embedding function uses a neural network language model.
17. The non-transitory computer-readable device of claim 15 , wherein the embedding function uses a Bidirectional Encoder Representations from Transformers (BERT) model.
18. The non-transitory computer-readable device of claim 15 , wherein the document file is an image of a document and the vector grid provides two-dimensional layout information of the image of the document.
19. The non-transitory computer-readable device of claim 15 , wherein analyzing the one or more symbols further comprises:
identifying a radius of neighboring symbols; and
selecting the one or more neighboring symbols within the radius.
20. The non-transitory computer-readable device of claim 15 , wherein the symbol grid includes a first symbol and a second symbol having a common symbol value and wherein a first contextual vector value corresponding to the first symbol differs from a second contextual vector value corresponding to the second symbol.