US 11,106,716 B2Granted

Automatic hierarchical classification and metadata identification of document using machine learning and fuzzy matching

AssigneeAccenture Global Solutions Limited

InventorsUrvesh Bhowan, Pedro Sacristan, Laura O'Malley, Abhilash Alexander Miranda, Medb Corcoran

Life of this patent

Estimating the term…

Loading patent…

Abstract

A hierarchical document classification system is disclosed. The system includes a text-based document classifier model for classifying an input electronic document into one of a set of predefined document categories. The system further includes an image-based metadata identification model for classifying electronic documents of a particular document category into a set of metadata categories. The system further includes a fuzzy text matcher for supplementing classification accuracy of the image-based metadata identification model to obtain a metadata category for the input electronic document.

Claims20

Dependent claims

A system comprising:

a database comprising metadata data elements organized in metadata fields, where the metadata fields include at least a metadata category field containing metadata category data elements belonging to a predetermined set of metadata categories;
a memory for storing a text-based document classifier model established based on a machine learning algorithm, an image-based metadata category identification model;
and a fuzzy text matcher;
and
system circuitry in communication with the database and the memory, the system circuitry configured to:
receive an electronic document;
generate a text document containing extractable text from the electronic document and an image representing a page view of the electronic document;
determine that the electronic document belongs to a predetermined set of document categories by using the text-based document classifier model to process the text document;
in response to determining that the electronic document belongs to the predetermined set of document categories:
obtain a first association of the image with a first metadata category data element of the metadata category field using the image-based metadata category identification model, wherein the image-based metadata category identification model is pretrained using a training corpus of images labeled using a subset of metadata categories of the predetermined set of metadata;
obtain a second association of the electronic document with a subset of the metadata data elements of the database by using the fuzzy text matcher to approximately match segments of the text document with the subset of the metadata data elements;
identify a subset of the metadata category data elements corresponding to the subset of the metadata data elements in the database;
select a single metadata category data element among a combination of the first metadata category data element and the subset of the metadata category data elements based on an identification probability for the first association and fuzzy matching probabilities for the second association;
and
attach the selected single metadata category data element as a metadata category label to the electronic document.

12 more elements

A method comprising:

receiving an electronic document by a system having a memory and system circuitry;
generating a text document containing extractable text from the electronic document using a text-based document classifier model established based on a machine learning algorithm and stored in the memory and generating an image representing a page view of the electronic document;
determining that the electronic document belongs to a predetermined set of document categories by using the text-based document classifier model to process the text document;
in response to determining that the electronic document belongs to the predetermined set of document categories:
obtaining a first association of the image with a first metadata category among a subset of metadata categories of a predetermined set of metadata categories using an image-based metadata category identification model stored in the memory, where the predetermined set of metadata categories are stored as data elements in a metadata category field in a database of the system containing metadata data elements organized in metadata fields including the metadata category field, and wherein the image-based metadata category identification model is pretrained using a training corpus of images labeled using the subset of metadata categories;
obtaining a second association of the electronic document with a subset of the metadata data elements of the database by using a fuzzy text matcher stored in the memory to approximately match segments of the text document with the subset of the metadata data elements;
identifying a subset of the predetermined set of metadata categories corresponding to the subset of the metadata data elements in the database;
selecting a single metadata category among a combination of the first metadata category and the subset of the predetermined set of metadata categories corresponding to the subset of the metadata data elements of the database as a label for the electronic document based on an identification probability for the first association and fuzzy matching probabilities for the second association;
and
attaching the selected single metadata category data element as a metadata category label to the electronic document.

7 more elements

Dependencies are inferred from the claim text.

Facts

Application: 15/810,885Filed 13 Nov 2017
Published: US 2019/0147103 A116 May 2019
Granted: 31 Aug 2021
Term ends: Not estimatedThe source records don’t support an estimate.

Classification15 codes

G06V 10/82
Using pattern recognition or machine learning: using neural networks
G06F 16/2468
Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries: fuzzy queries
G06F 16/353
Clustering; Classification: into predefined classes
G06F 16/93
Details of database functions independent of the retrieved data types: document management systems