# US 12,002,276 B2 · Document distinguishing based on page sequence learning

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,002,276 B2
- Document ID: US12002276B2
- Status: Granted
- Kind: B2
- Application number: 17/208,223
- Filed: 2021-03-22
- Published: 2022-09-22
- Granted: 2024-06-04
- Earliest claimed: 2021-03-22
- Assignees: BILL Operations, LLC
- Applicants: Bill.com, LLC
- Inventors: Xin Geng Kelly; Sangammohan Harimohan Singh; Derek Chan; Eitan Anzenberg
- Web page: https://patentgrep.com/patent/US12002276B2

## CPC

- G06V 30/416: Extracting the logical structure, e.g. chapters, sections or page numbers; Identifying elements of the document, e.g. authors
- G06F 16/355: Creation or modification of classes or clusters
- G06F 16/93: Document management systems
- G06N 3/049: Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
- G06N 20/00: Machine learning
- G06V 30/412: Layout analysis of documents structured with printed lines or input boxes, e.g. business forms or tables
- G06Q 30/04: Billing or invoicing
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

The accuracy of existing machine learning models, software technologies, and computers are improved by estimating whether a particular page belongs to a same document as another page or whether the page belongs to a different document. Such document distinguishing can be based on deriving relationship information between a first feature vector representing the page and a second feature vector representing the other page. This also improves the user experience and model building experience, among other things.

## Claims

### Claim 1

A computerized system comprising:
one or more processors; and
a non-transitory computer storage memory having computer-executable instructions stored thereon which, when executed by the one or more processors, implement a method comprising:
receiving, at a trained first machine learning model, image data of a multi-page document, the multi-page document including a first document that includes one or more first pages of the multi-page document and a second document that includes one or more second pages of the multi-page document;
in response to the receiving, at the trained first machine learning model, of the image data, generating a first feature vector embedding for the one or more first pages of the multi-page document and a second feature vector embedding for the one or more second pages of the multi-page document, the first and second vector embedding embedded based on learned patterns for document characteristics; and
based on feeding the first feature vector embedding and the second feature vector embedding to a second machine learning model determine whether each page, of the one or more first pages and the one or more second pages, is a continuation page of a previous page of a document or whether each page is a starting page of a new document; and
based on the determining, distinguishing the first document from the second document.

### Claim 2 (depends on 1)

The system of claim 1, wherein the first machine learning model is a computer-vision-based machine learning model.

### Claim 3 (depends on 1)

The system of claim 1, wherein the second machine learning model is one of: a Long-Short-Term-Memory (LSTM) model or a Bi-Directional LSTM.

### Claim 4 (depends on 1)

The system of claim 1, wherein the first document is a first invoice and the second document is a second invoice.

### Claim 5 (depends on 1)

The system of claim 1, wherein the method further comprises:
receiving, prior to the receiving of the image data, a second set of documents, the second set of documents representing user-identified documents;
extracting a set of features from the second set of documents;
training the second machine learning model based on learning weights associated with the set of features, wherein the determining is based at least in part on the training.

### Claim 6 (depends on 1)

The system of claim 1, wherein the determining is based on using a binary classification.

### Claim 7 (depends on 1)

The system of claim 1, wherein the determining is based at least in part on concatenating information from one or more previous pages to predict whether a currently analyzed page belongs to the first document or the second document.

### Claim 8

A computer-implemented method comprising:
receiving a plurality of documents, each document including one or more pages;
converting, via a first machine learning model, each page of the plurality of documents into a feature vector, the feature vector indicates one or more characteristics of the corresponding page;
based at least in part on the converting, deriving via a second machine learning model, relationship information between a first feature vector representing a first page and a second feature vector representing a second page, the first page being sequentially ordered first relative to the second page;
based at least in part on the deriving of the relationship information, determining that one of: the second page is a continuation of the first page of a first document or the second page is a starting page of a second document; and
based on the determining that one of: the second page is the continuation of the first page of the first document or the second page is a starting page of the second document, determining whether the second page belongs to a same document as the first page or whether the second page belongs to a different document, of the plurality of documents.

### Claim 9 (depends on 8)

The method of claim 8, wherein the first machine learning model is a computer-vision-based machine learning model.

### Claim 10 (depends on 8)

The method of claim 8, wherein the second machine learning model is one of: a Long-Short-Term-Memory (LSTM) model or a Bi-Directional LSTM.

### Claim 11 (depends on 8)

The method of claim 8, wherein the plurality of documents are invoice documents.

### Claim 12 (depends on 8)

The method of claim 8, further comprising:
receiving, prior to the receiving of the plurality of documents, a second set of documents, the second set of documents representing actual customer data;
extracting a set of features from the second set of documents;
training the second machine learning model based on learning weights associated with the set of features, wherein the determining whether the second page belongs to a same document as the first page or whether the second page belongs to a different document, of the plurality of documents is based at least in part on the training.

### Claim 13 (depends on 8)

The method of claim 8, wherein the determining that one of: the second page is a continuation of the first page of a first document or the second page is a starting page of a second document is based on using a binary classification.

### Claim 14 (depends on 8)

The method of claim 8, wherein the determining whether the second page belongs to a same document as the first page or whether the second page belongs to a different document, of the plurality of documents is based at least in part on concatenating information from one or more previously ordered pages and one or more later ordered pages to predict whether a currently analyzed page belongs to a first document or a second document.

### Claim 15 (depends on 8)

The method of claim 8, further comprising:
determining that the second page belongs to a same document as the first page; and
based at least in part on the determining, generating a user interface element.

### Claim 16 (depends on 8)

The method of claim 8, further comprising:
determining that the second page belongs to a different document relative to the a first document; and
based at least in part on the determining, generating a user interface element.

### Claim 17

One or more non-transitory computer storage media 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, each document including one or more pages;
converting, via a first machine learning model, one or more pages of the plurality of documents into a feature vector, the feature vector indicates one or more characteristics of the one or more pages;
based at least in part on the converting, deriving relationship information between a first feature vector representing a first page and a second feature vector representing a second page;
based at least in part on the deriving, determining that one of: the second page immediately follows the first page of a first document or the second page is a starting page of a second document; and
based at least in part on the determining that one of: the second page immediately follows the first page of the first document or the second page is a starting page of the second document, estimating whether the second page belong to a same document as the first page or whether the second page belongs to a different document, of the plurality of documents.

### Claim 18 (depends on 17)

The one or more non-transitory computer storage media of claim 17, wherein the first machine learning model is a deep neural network, and wherein the estimating includes using a classifier model, and wherein the deriving includes using one of: a Long-Short-Term-Memory (LSTM) model or a Bi-Directional LSTM.

### Claim 19 (depends on 17)

The one or more non-transitory computer storage media of claim 17, wherein the plurality of documents are invoice documents.

### Claim 20 (depends on 17)

The one or more non-transitory computer storage media of claim 17, wherein the estimating includes determining, using a binary classification, whether the second one or more pages is a continuation of the first one or more pages of a first document or whether the second one or more pages is a starting page of a second document.