US 12,283,363 B2
Web application for service recommendations with machine learning
Kelly Goncalves, Toronto (CA); Russell Goldman, Toronto (CA); Neda Paryab, Toronto (CA); Sidhant Kapahi, Toronto (CA); Maria Winslow, Toronto (CA); Chai Lam, Toronto (CA); and Hannah McIsaac, Toronto (CA)
Assigned to ROYAL BANK OF CANADA, Toronto (CA)
Filed by ROYAL BANK OF CANADA, Toronto (CA)
Filed on Aug. 12, 2021, as Appl. No. 17/401,066.
Claims priority of provisional application 63/073,786, filed on Sep. 2, 2020.
Prior Publication US 2022/0067814 A1, Mar. 3, 2022
Int. Cl. G16H 20/70 (2018.01); G06F 9/54 (2006.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01); G16H 50/30 (2018.01); G06Q 40/02 (2023.01)
CPC G16H 20/70 (2018.01) [G06F 9/541 (2013.01); G06N 7/01 (2023.01); G06N 20/00 (2019.01); G16H 50/30 (2018.01); G06Q 40/02 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A system comprising:
a memory storing a resource database, persona database, a user database, at least one machine learning model, and a risk model; and
a hardware processor for a web application having an interface with tools for resource recommendations and a next best action for a stepped-care model, the web application coupled to a plurality of microservices to exchange data to populate the tools of the interface with the resource recommendations, the interface configured to monitor electronic interactions to collect content interaction data;
the processor having a persona detection engine to compute a persona for a user using the at least one machine learning model, user data and the electronic interactions, wherein a persona is defined as a set of user preferences or attributes, wherein the persona detection engine implements one or more clustering processes to segment and cluster the user data to identify a set of clusters for personas and generate a representative persona by mapping the user preferences or attributes to dominating user preferences or attributes in the set of clusters for personas, wherein the persona detection engine implements content-to-content similarity measures for the electronic interactions to generate one or more user preferences or attributes of the set of user preferences or attributes for the persona, the processor having a risk model that uses one or more clustering processes to compute user preferences or attributes for likely claimants using financial attributes, the processor determining a high risk user based on the cluster of user preferences or attributes for likely claimants, wherein the persona detection engine compares the cluster of user preferences or attributes for likely claimants to the set of user preferences or attributes for the persona, the persona linked to preferred service types for the set of resources;
the processor having a personalization engine that interacts with the persona detection engine to generate a set of resources for the user for the resource recommendations by using a hybrid personalization inference to extract adaptive preferences based on the persona of the user and output of the at least one machine learning model, wherein the at least one machine learning model comprises a hybrid model of the content-to-content similarity to detect similarities in content from the resource database and user preferences to predict one or more resources that the user is likely to prefer for the set of resources, and collaborative filtering by detecting users with similar behaviours;
wherein the processor is configured to determine the next best action for the stepped-care model using the set of resources and the persona;
wherein the processor is configured to update at least one machine learning model using feedback from the interface.