US 12,694,204 B2Granted

System and method for comparing documents using an artificial intelligence (AI) model

AssigneeUST Global (Singapore) Pte. Ltd.

InventorsSandeep Sashi Kumar, Rajiv Shanbhag, Rajani Sreedharan

Life of this patent

Estimating the term…

Loading patent…

Abstract

A method for comparing documents using an AI model is disclosed herein. The method includes preprocessing a plurality of input documents into a structured, standardized format for comparison. Further, the method includes generating a layout of components in the plurality of input documents. Furthermore, the method includes generating a canonical representation indicating relationships between the components. Furthermore, the method includes determining sections within each of the plurality of input documents by segmenting the layout Furthermore, the method includes determining a textual difference matrix using the AI model, based on the comparison of the plurality of sections within each of the plurality of input documents, wherein the textual difference matrix indicates alterations in one or more expressions in each of the plurality of sections.

Claims12

Dependent claims

A method for comparing documents using an artificial intelligence (AI) model, the method comprising:

preprocessing a plurality of input documents into a structured, standardized format for comparison;
generating a layout indicative of an arrangement of one or more components in each of the plurality of input documents using the AI model, based on parsing the one or more components,
wherein generating the layout comprises:
predicting a positional coordinate corresponding to each of the one or more components for each of the plurality of input documents, wherein predicting the positional coordinate comprises:
training the AI model with an annotated plurality of training documents based on at least one of a precision value, a recall value, and a Mean Average Precision (mAP) value associated with the AI model;
determining hyperparameters associated with the annotated plurality of training documents based on the training;
learning the positional coordinate corresponding to the one or more components in the plurality of training documents based on the determined hyperparameters;
and
predicting the positional coordinate corresponding to the one or more components for each of the plurality of input documents based on the learning;
parsing each of the one or more components in each of the plurality of input documents based on the positional coordinate to each of the one or more components;
and
generating the layout for each of the plurality of input documents based on the parsing;
generating a canonical representation based on the layout, wherein the canonical representation indicates relationships between the one or more components of each of the plurality of input documents;
determining a plurality of sections within each of the plurality of input documents by segmenting the layout based on the canonical representation, wherein the plurality of sections indicates logical units in the corresponding input document of the plurality of input documents;
and
determining a textual difference matrix using the AI model, based on a comparison of the plurality of sections within each of the plurality of input documents, wherein the textual difference matrix indicates an alteration in one or more expressions in each of the plurality of sections.

13 more elements

A system for comparing documents using an artificial intelligence (AI) model, the system comprising:

a memory;
and
at least one processor in communication with the memory, wherein the at least one processor is configured to:
preprocess a plurality of input documents into a structured, standardized format for comparison;
generate a layout indicative of an arrangement of one or more components in each of the plurality of input documents using the AI model, based on parsing the one or more components,
wherein to generate the layout, the at least one processor is configured to:
predict a positional coordinate corresponding to each of the one or more components for each of the plurality of input documents, wherein to predict the positional coordinate, the at least one processor is configured to:
train the AI model with an annotated plurality of training documents based on at least one of a precision value, a recall value, and a Mean Average Precision (mAP) value associated with the AI model;
determine hyperparameters associated with the annotated plurality of training documents based on the training;
learn the positional coordinate corresponding to the one or more components in the plurality of training documents based on the determined hyperparameters;
and
predict the positional coordinate corresponding to the one or more components for each of the plurality of input documents based on the learning;
parse each of the one or more components in each of the plurality of input documents based on the positional coordinate to each of the one or more components;
and
generate the layout for each of the plurality of input documents based on the parsing;
generate a canonical representation based on the layout, wherein the canonical representation indicates relationships between the one or more components of each of the plurality of input documents;
determine a plurality of sections within each of the plurality of input documents by segmenting the layout based on the canonical representation, wherein the plurality of sections indicate logical units in the corresponding input document of the plurality of input documents;
and
determine a textual difference matrix using the AI model, based on a comparison of the plurality of sections within each of the plurality of input documents, wherein the textual difference matrix indicates an alteration in one or more expressions in each of the plurality of sections.

16 more elements

Dependencies are inferred from the claim text.