US-12620024-B2 - Method and system for predicting user intentions within a digital banking application

US12620024B2US 12620024 B2US12620024 B2US 12620024B2US-12620024-B2

Abstract

Methods for using and training a classifier to predicting user intentions as the user navigates in a digital banking application. A pathway dataset, excluding financial information, is acquired, capturing a sequence of navigation events during a user session. These navigation events, representing user interactions with the application interface, are converted into a sequence of latent space representations and input into a classifier. The classifier is trained using a historical dataset of navigation events annotated with intention labels as ground truth. Upon identifying a predicted user intention, the method suggests a corresponding service via the graphical user interface (GUI). This method ensures accurate intention prediction without accessing sensitive financial information, so that privacy and usability in digital banking environments are enhanced.

Inventors

Alex T. Lau
Juan Martin
Gautam DHALL
Neil MENDONCA
Charles Thomson
Arup Saha
Jigisha PATEL
Roop MALHANS
Yuanfu TIAN

Assignees

ROYAL BANK OF CANADA

Dates

Publication Date: 20260505
Application Date: 20241031

Claims (19)

1 . A method comprising: acquiring a pathway dataset excluding financial information as a user navigates in a digital banking application in a current user session, the pathway dataset comprising a time-ordered sequence of navigation events that have occurred in the current user session, wherein each of the navigation events is an interactive action by the user with an application interface of the digital banking application; converting the pathway dataset into a sequence of latent space representations; inputting the sequence of latent space representations to a classifier trained by: acquiring a training dataset excluding the financial information, the training dataset comprising a plurality of historical navigation events, wherein at least one of the plurality of historical navigation events is annotated with one of a plurality of intention labels as ground truth; converting the training dataset into at least one training sequence of latent space representations; and training the classifier using the at least one training sequence of latent space representations and the annotated one of the plurality of intention labels; and in response to the classifier identifying the sequence of latent space representations to be one of the plurality of intention labels, assigning the identified intention label as a predicted user intention; and suggesting a service corresponding to the predicted user intention via a graphic user interface (GUI) of the digital banking application.
2 . The method of claim 1 , wherein converting the pathway dataset further comprises vectorizing the sequence of latent space representations into a pathway feature vector, and wherein converting the training dataset comprises vectorizing the at least one training sequence of latent space representations into at least one training feature vector.
3 . The method of claim 1 , wherein the classifier identifies the sequence of latent space representations to be one of the plurality of intention labels in response to determining that a confidence level associated with the one of the plurality of intention labels exceeds a predefined confidence threshold.
4 . The method of claim 1 , further comprising recording, by a tracking service module, actions of the user as the user is navigating in the digital banking application in the current user session to generate the pathway dataset.
5 . The method of claim 1 , wherein the training dataset further comprises a plurality of historical identifiers of a screen of the GUI of the digital banking application, wherein at least one of the plurality of historical identifiers of the screen is annotated with one of the plurality of intention labels by the human.
6 . The method of claim 5 , wherein the pathway dataset further comprises a plurality of identifiers of the screen of the GUI that have occurred in the current user session.
7 . The method of claim 1 , wherein the plurality of intention labels comprise at least one of paying bills, transferring funds, or email money transfer.
8 . The method of claim 1 , wherein the training dataset is collected from a plurality of historical users, and wherein, for each of the plurality of historical users, the plurality of historical navigation events are from different user sessions of the digital banking application.
9 . The method of claim 8 , wherein the user is one of the plurality of historical users.
10 . The method of claim 1 , wherein the classifier has a feedforward neural network architecture characterized by five distinct hidden layers.
11 . The method of claim 10 , wherein the five hidden layers in sequence host 100, 250, 200, 100, and 50 neurons, respectively.
12 . The method of claim 1 , wherein the classifier has a convolutional neural network architecture comprising three 1-dimensional convolutional layers.
13 . The method of claim 12 , wherein the three 1-dimensional convolutional layers comprise 32 filters, 128 filters, and 128 filters, respectively, and wherein the three 1-dimensional convolutional layers have a kernel size of 3, 5, and 5, respectively.
14 . The method of claim 1 , wherein converting the training dataset further comprises: in response to the historical navigation events in the training dataset that precipitate a first historical navigation event annotated with one of a plurality of intention labels being fewer than a predetermined number, padding the training dataset to meet the predetermined number; and converting the predetermined number of historical navigation events into the sequence of latent space representations.
15 . The method of claim 14 , wherein the predetermined number ranges from 30 to 70.
16 . The method of claim 1 , wherein at least one of the conversion of the training dataset or the conversion of the pathway dataset is performed using a look-up table or an encoder.
17 . The method of claim 1 , wherein the training dataset is acquired from a web analytics service used to monitor the digital banking application.
18 . The method of claim 1 , wherein the classifier has an output layer characterized by three neurons.
19 . A non-transitory computer readable medium have stored thereon computer program code that is executable by at least one processor and that, when executed by the at least one processor, causes the at least one processor to perform a method comprising: acquiring a pathway dataset excluding financial information as a user navigates in a digital banking application in a current user session, the pathway dataset comprising a time-ordered sequence of navigation events that have occurred in the current user session, wherein each of the navigation events is an interactive action by the user with an application interface of the digital banking application; converting the pathway dataset into a sequence of latent space representations; inputting the sequence of latent space representations to a classifier trained by: acquiring a training dataset excluding the financial information, the training dataset comprising a plurality of historical navigation events, wherein at least one of the plurality of historical navigation events is annotated with one of a plurality of intention labels as ground truth; converting the training dataset into at least one training sequence of latent space representations; and training the classifier using the at least one training sequence of latent space representations and the annotated one of the plurality of intention labels; and in response to the classifier identifying the sequence of latent space representations to be one of the plurality of intention labels, assigning the identified intention label as a predicted user intention; and suggesting a service corresponding to the predicted user intention via a graphic user interface (GUI) of the digital banking application.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS The present applications claims priority to U.S. provisional patent application No. 63/594,515, filed on Oct. 31, 2023, and entitled, “Method and System for Predicting User Intentions within a Digital Banking Application,” the entirety of which is hereby incorporated by reference herein. TECHNICAL FIELD The present disclosure relates to methods, systems, and techniques for predicting user intentions in a digital banking application. More particularly, the present disclosure relates to methods, systems, and techniques for predicting user intentions using navigation information and neural network(s), without accessing the user's sensitive financial data such as transaction details, account balances, and withdrawal amounts. BACKGROUND The digital landscape, particularly in the banking sector, has undergone significant changes in recent years. As consumers increasingly turn to mobile and online platforms for their financial transactions, digital banking applications have become an essential touchpoint between financial institutions and their customers. The ubiquity of smartphones and the convenience of digital transactions have led to a tremendous increase in mobile-first transactions in the banking industry. In this era of digital transformation, the user experience (UX) of mobile applications plays a critical role in determining customer satisfaction and retention rates. A seamless and intuitive UX not only ensures smooth transactions, but also builds trust and confidence in the digital platform. Historically, banking applications have focused primarily on providing secure and efficient transaction capabilities. However, with the evolving expectations of the modern user, there is a growing emphasis on personalizing the user experience based on individual preferences and behaviors. SUMMARY According to a first aspect, there is provided a method comprising: acquiring a pathway dataset excluding financial information as a user navigates in a digital banking application in a current user session, the pathway dataset comprising a time-ordered sequence of navigation events that have occurred in the current user session, wherein each of the navigation events is an interactive action by the user with an application interface of the digital banking application; converting the pathway dataset into a sequence of latent space representations; inputting the sequence of latent space representations to a classifier; in response to the classifier identifying the sequence of latent space representations to be one of the plurality of intention labels, assigning the identified intention label as a predicted user intention; and suggesting a service corresponding to the predicted user intention via a graphic user interface (GUI) of the digital banking application. The classifier is trained by acquiring a training dataset excluding the financial information, the training dataset comprising a plurality of historical navigation events, wherein at least one of the plurality of historical navigation events is annotated with one of a plurality of intention labels as ground truth; converting the training dataset into at least one training sequence of latent space representations; and training the classifier using the at least one training sequence of latent space representations and the annotated one of the plurality of intention labels. In some embodiments, converting the pathway dataset may further comprise vectorizing the sequence of latent space representations into a pathway feature vector, and wherein converting the training dataset may comprise vectorizing the at least one training sequence of latent space representations into at least one training feature vector. In some embodiments, the classifier may identify the sequence of latent space representations to be one of the plurality of intention labels in response to determining that a confidence level associated with the one of the plurality of intention labels exceeds a predefined confidence threshold. In some embodiments, the method may further comprise recording, by a tracking service module, actions of the user as the user is navigating in the digital banking application in the current user session to generate the pathway dataset. In some embodiments, the training dataset may further comprise a plurality of historical identifiers of a screen of the GUI of the digital banking application, wherein at least one of the plurality of historical identifiers of the screen is annotated with one of the plurality of intention labels by the human. In some embodiments, the pathway dataset may further comprise a plurality of identifiers of the screen of the GUI that have occurred in the current user session. In some embodiments, the plurality of intention labels may comprise at least one of paying bills, transferring funds, or email money transfer. In some embodiments, the training dataset may be collected from a plurality of historical users, and w