Back to results
US 12,046,063 B2Granted
Table information extraction and mapping to other documents
AssigneeBILL Operations, LLC
InventorsNatalia Berestovsky, Stefano Andrea Romano, Ricardo Antonio Fernandez, Joseph Michael Price
Life of this patent
Estimating the term…
Abstract
The accuracy of existing machine learning models, software technologies, and computers are improved by using one or more machine learning models to map data inside structural elements, such as rows or columns, as found within a document to data objects of other documents, where the data objects are at least partially indicative of candidate categories that the data can belong to.
Claims20
A computerized system comprising:
- one or more processors;
- and
- computer storage memory having computer-executable instructions stored thereon which, when executed by the one or more processors, implement a method comprising:
- receiving a first document associated with a user;
- generating a set of bounding coordinates that define a table within the first document;
- in response to the generating of the set of bounding coordinates that define the table, detecting a plurality of rows from the table, each row comprising a numerical value;
- in response to the detecting of the plurality of rows, encoding a first one or more words in a first detected row, of the plurality of detected rows, into a first feature vector;
- computing a first distance between the first feature vector and a second feature vector, the second feature vector represents a data object that at least partially indicates, at a second document associated with the user, a candidate category that the first one or more words belong to, the one or more words describing an item and the candidate category describing multiple items;
- encoding a second one or more words in a second detected row, of the plurality of rows, into a third feature vector;
- computing a second distance between the third feature vector and the second feature vector,
- based at least in part on the first distance and the second distance being within a threshold, performing a mathematical operation on the numerical value from the first row and the second row;
- and
- based at least in part on the performing of the mathematical operation on the numerical value from the first row and the second row, generating a third document that includes the data object and an aggregated numerical value that is a result of the performing of the mathematical operation.
10 more elements
A computer-implemented method comprising:
- receiving a first document associated with a user;
- determining a table within the first document;
- extracting a plurality of structural elements from the table, a first structural element comprising a first cost of a first good and a second structural element comprising a second cost of a second good;
- determining a first similarity score between a first set of words and a data object, the first set of words describing the first good and being included in the first structural element, the data object at least partially indicates, at a second document associated with the user, a first candidate category that the first set of words belong to, the candidate category not describing the first good but describing a good type that the first good belongs to;
- determining a second similarity score between a second set of words and the data object, the second set of words describing the second good and being included in the second structural element;
- based at least in part on the first similarity score meeting a similarity threshold and the second similarity score not meeting the similarity threshold, applying an operation on the first cost but not the second cost resulting in an output cost;
- and
- based at least in part on the applying of the operation, generating a third document that includes the data object and the output cost.
5 more elements
One or more non-transitory computer readable storage mediums having computer-executable instructions embodied thereon that, when executed, by one or more processors, cause the one or more processors to perform a method, the method comprising:
- receiving a plurality of documents associated with a user, each document describes, via one or more identifiers, one or more categories of item expenses;
- subsequent to the receiving, receiving a first document associated with the user, the first document not being among the plurality of documents;
- determining a table within the first document, wherein the table describes individual expenses of multiple goods;
- extracting a plurality of structural elements from the table, a first structural element comprising a first cost of a first good and a second structural element comprising a second cost of a second good;
- determining a first similarity score between a first set of words and a data object, the first set of words describe the first good and being included in the first structural element, the data object at least partially indicates, a candidate category that the first set of words belong to, the candidate category not describing the first good but describing a good type that the first good belongs to;
- based at least in part on the first similarity score meeting a similarity threshold, applying an operation on the first cost resulting in an output cost;
- and
- based at least in part on the applying of the operation, generating a third document that includes the data object and the output costs.
5 more elements
Dependencies are inferred from the claim text.