Method and system for user experience personalization in data management systems using machine learning

Patent metadata

Field	Value
Publication number	US-11269665-B1
Application number	US-201815938168-A
Country	US
Kind code	B1
Filing date	Mar 28, 2018
Priority date	Mar 28, 2018
Publication date	Mar 8, 2022
Grant date	Mar 8, 2022

Abstract

A method and system provides personalized assistance to users of a data management system. The method and system trains an analysis model with both a supervised machine learning process and an unsupervised machine learning process to identify relevant assistance topics based on a user query and the attributes of the user that provided the query. The method and system outputs personalized assistance to the user based on the analysis of the analysis model.

First claim

What is claimed is:

1. A method performed by one or more processors of a system, the method comprising:
training, with a supervised machine learning process, a first sub-model of an analysis model to identify relevant assistance topics for users of a tax return preparation system;
training, with an unsupervised machine learning process, a second sub-model of the analysis model to identify relevant assistance topics for the users of the tax return preparation system;
training, with an unsupervised deep learning machine learning process, a third sub-model of the analysis model to identify relevant assistance topics for the users of the tax return preparation system;
providing tax return preparation interview content to a current user of the tax return preparation system as part of a tax return preparation interview;
receiving a query from the current user corresponding to a request for assistance related to the tax return preparation system, wherein the query includes a query provided to a conversational interface of the tax return preparation system;
retrieving attributes of the current user including at least user profile data indicating one or more groups of historical users of the tax return preparation system that are similar to the current user, and tax related data of the current user;
analyzing, with the first sub-model of the analysis model, the query and the attributes based on the supervised machine learning process by using a multiclass classifier that identifies one or more assistance topics relevant to the current user based on the query and the attributes;
analyzing, with the second sub-model of the analysis model, the query and the attributes based on the unsupervised machine learning process by:
generating a first augmented query by combining additional query terms based on the attributes with the query; and
identifying assistance topics associated with the first augmented query by using a Latent Dirichlet Allocation process;

analyzing, with the third sub-model of the analysis model, the query and the attributes based on the unsupervised deep learning machine learning process, by:
generating a second augmented query by combining the additional query terms based on the attributes with text of the query;
generating vectorization data by converting each term from the second augmented query to a numerical vector; and
identifying assistance topics associated with the second augmented query by applying a grouping algorithm to the vectorization data;

generating personalized assistance data based on the analyses by the first, second, and third sub-models, and by:
selecting a search engine from a plurality of search engines based on the assistance topics; and
searching, with the selected search engine, among a database of assistance documents based on the assistance topics; and

outputting the personalized assistance data to the current user based on the assistance topics.

2. The method of claim 1 , wherein the attributes further include one or more of clickstream data indicating tax return preparation screens that have been visited by the current user in the tax return preparation interview, demographics data associated with the current user, or tax return preparation data received from the current user in association with the tax return preparation interview.

3. The method of claim 1 , wherein generating the vectorization data includes applying one or more of a word2vec process, a doc2vec process, or a sentence2vec process.

4. The method of claim 1 , wherein the grouping algorithm includes a K nearest neighbors clustering algorithm.

5. The method of claim 1 , wherein the supervised machine learning process includes training the first sub-model with training set data including attributes of the historical users and how the historical users selected assistance documents from search results.

6. A system comprising:
at least one processor; and
at least one memory storing instructions that, when executed by the at least one processor, cause the system to:
train, with a supervised machine learning process, a first sub-model of an analysis model to identify relevant assistance topics for users of a tax return preparation system;
train, with an unsupervised machine learning process, a second sub-model of the analysis model to identify relevant assistance topics for the users of the tax return preparation system;
train, with an unsupervised deep learning machine learning process, a third sub-model of the analysis model to identify relevant assistance topics for the users of the tax return preparation system;
provide tax return preparation interview content to a current user of the tax return preparation system as part of a tax return preparation interview;
receive a query from the current user corresponding to a request for assistance related to the tax return preparation system, wherein the query includes a query provided to a conversational interface of the tax return preparation system;
retrieve attributes of the current user including at least user profile data indicating one or more groups of historical users of the tax return preparation system that are similar to the current user, and tax related data of the current user;
analyze, with the first sub-model of the analysis model, the query and the attributes based on the supervised machine learning process by using a multiclass classifier that identifies one or more assistance topics relevant to the current user based on the query and the attributes;
analyze, with the second sub-model of the analysis model, the query and the attributes based on the unsupervised machine learning process by:
generating a first augmented query by combining additional query terms based on the attributes with the query; and
identifying assistance topics associated with the first augmented query by using a Latent Dirichlet Allocation process;

analyze, with the third sub-model of the analysis model, the query and the attributes based on the unsupervised deep learning machine learning process, by:
generating a second augmented query by combining the additional query terms based on the attributes with text of the query;
generating vectorization data by converting each term from the second augmented query to a numerical vector; and
identifying assistance topics associated with the second augmented query by applying a grouping algorithm to the vectorization data;

generate personalized assistance data based on the analyses by the first second, and third sub-models, and by:
selecting a search engine from a plurality of search engines based on the assistance topics; and
searching, with the selected search engine, among a database of assistance documents based on the assistance topics; and

output the personalized assistance data to the current user based on the assistance topics.

7. The system of claim 6 , wherein the attributes further include one or more of clickstream data indicating tax return preparation screens that have been visited by the current user in the tax return preparation interview, demographics data associated with the current user, or tax return preparation data received from the current user in association with the tax return preparation interview.

8. The system of claim 6 , wherein generating the personalized assistance data includes:
selecting, from a plurality of conversational interface agents, a conversational interface agent to respond to the query via the conversational interface based on the analyses by the first, second, and third sub-models; and
formulating a reply to the query with the selected conversational interface agent.

9. The system of claim 8 , wherein outputting the personalized assistance data to the current user includes

Assignees

• Intuit Inc

Inventors

• Podgorny Igor A
• Indyk Benjamin
• Sharafi Faraz
• Cannon Matthew

Classifications

• G06N3/02

  Neural networks · CPC title

• G06N20/00

  Machine learning · CPC title

• G06F9/453 Primary

  Help systems · CPC title

• G06F16/3347

  using vector based model · CPC title

• G06F16/3334 Primary

  Selection or weighting of terms from queries, including natural language queries · CPC title