Document term recognition and analytics

Patent metadata

Field	Value
Publication number	US-12380522-B2
Application number	US-202318328881-A
Country	US
Kind code	B2
Filing date	Jun 5, 2023
Priority date	Jan 10, 2019
Publication date	Aug 5, 2025
Grant date	Aug 5, 2025

Abstract

A device receives image data of a contractual document that includes an offer including terms of a proposed transaction, converts the image data to text data that identifies text within the contractual document, and receives preferences information for a recipient of the offer. The device identifies key terms within the contractual document by using term identification to analyze the text. The key terms may include a first key term that identifies subject matter of the proposed transaction and other key terms that are part of the offer. The device determines term scores that correspond to likelihoods of the other key terms being favorable to the recipient by using a data model to analyze the key terms and the preferences information. The device, based on the term scores, generates and provides another device with a recommendation to be used in determining whether the accept the offer.

First claim

What is claimed is:

1. A method, comprising:
training, by a device, a data model using a convolutional neural network (CNN) to receive a plurality of terms to generate a plurality of term scores,
wherein the plurality of terms are in a document, and
wherein training the data model comprises:
analyzing a set of vectors using a convolutional layer of the CNN by applying a filter to one or more portions of the set of vectors to create a feature map, and
analyzing the feature map, using the convolutional neural network, to output a plurality of M-dimensional vectors that indicate the plurality of term scores;

determining, by the device and using the data model, one or more unfavorable terms, from a set of terms, having unfavorable term scores from a set of term scores,
wherein the set of terms are from the plurality of terms,
wherein the set of term scores are from the plurality of term scores,
wherein the set of term scores correspond to one or more likelihoods of whether one or more terms, of the set of terms, are favorable or unfavorable based on user profile data, and
wherein the unfavorable term scores are less than a threshold,
wherein the threshold is based on the user profile data;

running, by the device, one or more simulations with one or more proposed modifications of one or more unfavorable term scores, from the set of term scores, that are less than the threshold, by modifying a value of the one or more unfavorable terms; and
providing, by the device, a recommendation regarding the document based on modifying the one or more unfavorable terms.

2. The method of claim 1 , wherein determining the set of term scores comprises:
providing the set of terms as input to the data model to cause the data model to output the set of term scores,
wherein the one or more terms identify purchasing information associated with the document, and
wherein one or more term scores, of the set of term scores, are based on whether the one or more terms that identify the purchasing information have values that are within one or more threshold ranges of values associated with corresponding terms that are found in the user profile data.

3. The method of claim 1 , wherein providing the recommendation comprises:
providing the recommendation based on indicating a weighted average of the set of term scores.

4. The method of claim 1 , wherein generating the recommendation comprises:
providing the recommendation to include the one or more proposed modifications to the one or more terms,
wherein the one or more proposed modifications include at least one of:
a first modification to add a new term to the document,
a second modification to remove a term, of the one or more terms, from the document, or
a third modification to change a particular term of the one or more terms in the document.

5. The method of claim 1 , wherein the document includes an offer with terms of a proposed transaction.

6. The method of claim 1 , further comprising:
identifying the set of terms within the document using a term matching technique to compare text included within the document and a master set of terms from the user profile data.

7. The method of claim 1 , wherein the document is a proposed contract with an offer, and
wherein providing the recommendation comprises:
providing the recommendation for display via another device to facilitate:
making a counteroffer, or
rejecting the offer.

8. A device, comprising:
one or more memories; and
one or more processors configured to:
train a data model using a convolutional neural network (CNN) to receive a plurality of terms to generate a plurality of term scores by:
analyzing a set of vectors using a convolutional layer of the CNN by applying a filter to one or more portions of the set of vectors to create a map, and
analyzing the map, using the convolutional neural network, to output a plurality of vectors that indicate the plurality of term scores;

determine one or more unfavorable terms, from a set of terms, having unfavorable term scores from a set of term scores,
wherein the set of terms are from the plurality of terms in a document,
wherein the set of term scores are from the plurality of term scores,
wherein the set of term scores correspond to one or more likelihoods of whether one or more terms are favorable or unfavorable based on user profile data, and
wherein the unfavorable term scores are less than a threshold,
wherein the threshold is based on the user profile data;

run one or more simulations with one or more proposed modifications of one or more unfavorable term scores, from the set of term scores, that are less than the threshold, by modifying a value of the one or more unfavorable terms; and
provide a recommendation regarding the document based on modifying the one or more unfavorable terms.

9. The device of claim 8 , wherein the user profile data includes one or more of:
data identifying past documents that were accepted,
data identifying past documents that were rejected,
purchasing preferences,
product preferences,
service preferences,
brand preferences, or
financial data.

10. The device of claim 8 , wherein the one or more processors, to provide the recommendation, are configured to:
provide the recommendation based on indicating a weighted average of the set of term scores.

11. The device of claim 8 , wherein the one or more terms include at least one of:
a term that identifies a user of the user profile data,
a term that identifies a maker of an offer, or
a term that identifies a subject matter of the document.

12. The device of claim 8 , wherein the one or more processors, to determine the set of term scores, are configured to:
provide the set of terms as input to the data model to cause the data model to output the set of term scores,
wherein the one or more terms identify purchasing information associated with the document, and
wherein one or more term scores, of the set of term scores, are based on whether the one or more terms that identify the purchasing information have values that are within one or more threshold ranges of values associated with corresponding terms that are found in the user profile data.

13. The device of claim 8 , wherein the one or more processors are further configured to:
identify a plurality of terms included in the document by analyzing text from the document using a tokenization technique;
compare the plurality of terms and a configured set of tokens; and
identify a subset of the plurality of terms, as the set of terms, based on the set of terms satisfying a threshold level of similarity with the configured set of tokens.

14. The device of claim 8 , wherein the one or more processors, to generate the recommendation, are to:
identify a new term to replace an unfavorable key term of the one or more unfavorable terms by analyzing other documents for particular subject matter that is similar to subject matter of the document.

15. A non-transitory computer-readable medium storing instructions, the instructions comprising:
one or more instructions that, when executed by one or more processors of a device, cause the one or more processors to:
train a data model using a convolutional neural network (CNN) to receive a plurality of terms to generate a plurality of term scores by:
analyzing a set of vectors using a convolutional layer of the CNN by applying a filter to one or more portions of the set of vectors to create a feature map, and
analyzing the feature map, using the convolutional neural network, to output a plurality of M-dimensional vectors that in

Assignees

• Capital One Services Llc

Inventors

• Awoyemi Olalekan
• Hoover Jason
• Duckworth Staevan
• Dagley Geoffrey
• Wylie Stephen
• Tang Qiaochu
• Price Micah

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06V30/413

  Classification of content, e.g. text, photographs or tables · CPC title

• G06F16/313

  Selection or weighting of terms for indexing · CPC title

• G06N20/00

  Machine learning · CPC title