Linking key-value pairs in documents

One or more non-transitory computer readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:

• generating, by processing a digital representation of a document including one or more regions of text, a set of feature maps that represent the one or more regions of text;
• resizing the set of feature maps based on sizes corresponding to the one or more regions of text;
• extracting one or more key-value pairs from the one or more regions of text based on the resized set of feature maps;
• generating, via execution of one or more machine learning models, a first set of bounding boxes for a first set of values associated with a first key within the one or more key-value pairs and a second set of bounding boxes for a second set of values associated with a second key within the one or more key-value pairs;
• generating a first set of mappings between a list of items included in the document and the first set of bounding boxes;
• generating a second set of mappings between the first set of bounding boxes and the second set of bounding boxes based on maximizing one or more of vertical column alignment or horizontal row alignment parameters between a first set of locations of the first set of bounding boxes and a second set of locations of the second set of bounding boxes, wherein the first set of locations and the second set of locations are non-overlapping;
• determining, for a given item included in the list of items, one or more associated bounding boxes included in the first set of bounding boxes and the second set of bounding boxes based on the first set of mappings and the second set of mappings;
• and
• extracting, for each given item included in the list of items, corresponding text included in the document based on the one or more associated bounding boxes.

A system, comprising:

• a memory that stores instructions, and
• a processor that is coupled to the memory and, when executing the instructions, is configured to:
• generate, by processing a digital representation of a document including one or more regions of text, a set of feature maps that represent the one or more regions of text;
• resize the set of feature maps based on sizes corresponding to the one or more regions of text;
• extract one or more key-value pairs from the one or more regions of text based on the resized set of feature maps;
• generate, via execution of one or more machine learning models, a first set of bounding boxes for a first set of values associated with a first key within the one or more key-value pairs and a second set of bounding boxes for a second set of values associated with a second key within the one or more key-value pairs;
• generate a first set of mappings between a list of items included in the document and the first set of bounding boxes;
• generate a second set of mappings between the first set of bounding boxes and the second set of bounding boxes based on maximizing one or more of vertical column alignment or horizontal row alignment parameters between a first set of locations of the first set of bounding boxes and a second set of locations of the second set of bounding boxes, wherein the first set of locations and the second set of locations are non-overlapping;
• determine, for a given item included in the list of items, one or more associated bounding boxes included in the first set of bounding boxes and the second set of bounding boxes based on the first set of mappings and the second set of mappings;
• and
• extract, for each given item included in the list of items, corresponding text included in the document based on the one or more associated bounding boxes.