US 10,614,345 B1Granted

Machine learning based extraction of partition objects from electronic documents

AssigneeErnst & Young U.S. LLP

InventorsDan G. Tecuci, Ravi Kiran Reddy Palla, Hamid Reza Motahari Nezhad, Vincent Poon, Nigel Paul Duffy, Joseph Nipko

Life of this patent

Estimating the term…

patentgrep

Loading patent…

Abstract

An object-extraction method includes generating multiple partition objects based on an electronic document, and receiving a first user selection of a data element via a user interface of a compute device. In response to the first user selection, and using a machine learning model, a first subset of partition objects from the multiple partition objects is detected and displayed via the user interface. A user interaction, via the user interface, with one of the partition objects is detected, and in response, a weight of the machine learning model is modified, to produce a modified machine learning model. A second user selection of the data element is received via the user interface, and in response and using the modified machine learning model, a second subset of partition objects from the multiple partition objects is detected and displayed via the user interface, the second subset different from the first subset.

Claims16

Dependent claims

A method, comprising:

generating, for an electronic document, a plurality of partition objects based on and associated with the electronic document;
receiving, via a user interface of a compute device, a first signal representing a user-selected data element of interest for the electronic document;
detecting, in response to receiving the first signal representing the user-selected data element and using a machine learning model, a first subset of partition objects from the plurality of partition objects based on a relevance of the first subset of partition objects to the user-selected data element of interest, the machine learning model including a semantic model;
displaying, via the user interface, a representation of each partition object from the first subset of partition objects;
detecting a user interaction, via the user interface, with a representation of a partition object from the first subset of partition objects, the user interaction representing a confirmation of relevance of the partition object from the first subset of partition objects;
in response to detecting the user interaction, modifying the machine learning model by adjusting a weight of the machine learning model based on the user interaction, to produce a modified machine learning model;
receiving, via the user interface of the compute device, a second signal representing the user-selected data element;
detecting, in response to receiving the second signal representing the user-selected data element and using the modified machine learning model, a second subset of partition objects from the plurality of partition objects, the second subset of partition objects different from the first subset of partition objects;
displaying, via the user interface, a representation of each partition object from the second subset of partition objects;
detecting that a number of user interactions associated with the user-selected data element exceeds a predefined threshold number of interactions;
in response to detecting that the number of user interactions associated with the user-selected data element exceeds the predefined threshold number of interactions, replacing the semantic model with a similarity model;
and
subsequently replacing the similarity model with a multilayer perceptron (MLP) model.

10 more elements

A method, comprising:

generating, for an electronic document, a plurality of partition objects based on and associated with the electronic document;
receiving, via a user interface of a compute device, a first signal representing a user-selected data element of interest for the electronic document;
detecting, in response to receiving the first signal representing the user-selected data element and using a machine learning model, a first subset of partition objects from the plurality of partition objects based on a relevance of the first subset of partition objects to the user-selected data element of interest, the machine learning model including a multilayer perceptron (MLP) model;
displaying, via the user interface, a representation of each partition object from the first subset of partition objects;
detecting a user interaction, via the user interface, with a representation of a partition object from the first subset of partition objects, the user interaction representing a confirmation of relevance of the partition object from the first subset of partition objects;
in response to detecting the user interaction, modifying the machine learning model by adjusting a weight of the machine learning model based on the user interaction, to produce a modified machine learning model;
receiving, via the user interface of the compute device;
a second signal representing the user-selected data element;
detecting, in response to receiving the second signal representing the user-selected data element and using the modified machine learning model, a second subset of partition objects from the plurality of partition objects, the second subset of partition objects different from the first subset of partition objects;
displaying, via the user interface, a representation of each partition object from the second subset of partition objects;
detecting that a number of user interactions associated with the user-selected data element exceeds a predefined threshold number of interactions;
in response to detecting that the number of user interactions associated with the user-selected data element exceeds the predefined threshold number of interactions, updating the MLP model.

9 more elements

A method, comprising:

generating, for an electronic document, a plurality of partition objects based on and associated with the electronic document;
receiving, via a user interface of a compute device, a first signal representing a user-selected data element of interest for the electronic document;
detecting, in response to receiving the first signal representing the user-selected data element and using a machine learning model, a first subset of partition objects from the plurality of partition objects based on a relevance of the first subset of partition objects to the user-selected data element of interest;
displaying, via the user interface, a representation of each partition object from the first subset of partition objects;
detecting a user interaction, via the user interface, with a representation of a partition object from the first subset of partition objects, the user interaction representing a confirmation of relevance of the partition object from the first subset of partition objects;
in response to detecting the user interaction, modifying the machine learning model by adjusting a weight of the machine learning model based on the user interaction, to produce a modified machine learning model;
receiving, via the user interface of the compute device, a second signal representing the user-selected data element;
detecting, in response to receiving the second signal representing the user-selected data element and using the modified machine learning model, a second subset of partition objects from the plurality of partition objects, the second subset of partition objects different from the first subset of partition objects;
displaying, via the user interface, a representation of each partition object from the second subset of partition objects;
and
detecting a cumulative number of user interactions, including the user interaction, associated with the user-selected data element,
the machine learning model being selected based on the cumulative number of user interactions associated with the user-selected data element.

9 more elements

Dependencies are inferred from the claim text.

Facts

Application: 16/382,707Filed 12 Apr 2019
Granted: 7 Apr 2020
Term ends: Not estimatedThe source records don’t support an estimate.

Classification25 codes

G06F 3/0482
Based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like…:…
G06F 18/2113
Selection of the most significant subset of features: by ranking or filtering the set of features, e.g. using a measure of variance or of feature…
G06F 18/2148
Generating training patterns; Bootstrap methods, e.g. bagging or boosting: characterised by the process organisation or structure, e.g. boosting…
G06F 18/2178
Validation; Performance evaluation; Active pattern learning techniques: based on feedback of a supervisor