Method and system for presenting items in online environment based on previous item selections

Patent metadata

Field	Value
Publication number	US-11769194-B2
Application number	US-201816160727-A
Country	US
Kind code	B2
Filing date	Oct 15, 2018
Priority date	Jun 18, 2018
Publication date	Sep 26, 2023
Grant date	Sep 26, 2023

Abstract

Methods and systems for predicting relevant items to be presented to a user in an online environment are described. The methods and systems described herein generate models based on previous item selections to determine an overall time series model for predicting a relevant time of next item selection as well as items most likely to be selected at that time. Complementary items can be presented to the user alongside the selection of most relevant items.

First claim

The invention claimed is:

1. A method of generating an item recommendation to a retail customer, the method comprising:
identifying sales data associated with a retail customer based on historical sales being associated with an identifier of the retail customer, the sales data including recurring sales data representing sales of items purchased a plurality of times within a predefined time period;
classifying the sales data into item level data, wherein the item level data includes frequency and timing of the retail customer's historical purchase of items;
for each item, identifying an item category;
for each retail customer and each item category, building a time-series model of purchases;
determining, from the time-series model of purchases for the retail customer and item category, one or more model features;
generating a plurality of time-series models from the one or more model features for the retail customer, the plurality of time-series models including a date model generated using an autoregressive integrated moving average (ARIMA) model and a rate model generated using a second ARIMA model, the rate model being independent of the date model;
performing a survival analysis on a per-item basis across a plurality of retail customers, the plurality of retail customers including the retail customer, individually for the retail customer, to determine a repurchase probability of an item within each item category, the repurchase probability being determined within a repurchase period from a cumulative hazard function;
at a Gradient-Boosted Tree-based ensemble model, combining predictions from at least the date model, the rate model, and the survival analysis, to generate a time series prediction of a next retail customer purchase date and items purchased on the date of the next retail customer purchase;
receiving a request generated from a retail website to generate a basket prediction for the retail customer;
based on a date at which the request to generate the basket prediction is received, identifying highest-likelihood items to be purchased based on the ensemble model prediction on the date at which the request is received, the highest-likelihood items being selected from across a plurality of item categories;
using the identified highest-likelihood items to be purchased, generating and displaying a basket prediction matrix for the retail customer, the basket prediction matrix being a two-dimensional matrix having a plurality of rows comprising ranked item categories and a plurality of columns comprising a plurality of items within each category ranked based at least in part on frequency and timing of the retail customer's historical purchases;
in response to generating the basket prediction matrix, outputting in real-time a basket prediction for the retail customer, the basket prediction including the highest ranking item in each item category within the basket prediction matrix;
determining a format of a user interface displayable on a device associated with the retail customer, wherein the format includes a size of a display screen;
based on the determined size of the user interface, determining a maximum number of items and a maximum number of item categories to be displayed in a carousel;
in response to the determined size of the user interface, adapting the carousel to display the determined maximum number of items, wherein the carousel is configured to display items across the plurality of item categories in response to determining the maximum number of items and the maximum number of item categories;
generating at least a portion of the user interface including the carousel, the carousel including the determined maximum number of items, wherein each item displayed in the carousel is from the basket prediction on the date at which the request to generate a basket prediction is received;
receiving a selection from the retail customer through the user interface, of at least one of the items from the basket prediction via the carousel and adding the selected at least one item to a basket for checkout; and
based, at least in part, on receiving the selection of the at least one item through the user interface to add to the basket for checkout, updating the time-series model of purchases for the retail customer.

2. The method of claim 1 , wherein the survival analysis applies a survival function based on likelihood of repurchase of an item within the item category for each of the plurality of retail customers.

3. The method of claim 1 , wherein the survival analysis for a particular item is performed based on a plurality of features including a daily survival rate for the particular item, a weekly survival rate for the particular item, a number of trips made by the retail customer since last purchase of the particular item, a number of days since the retail customer has purchased the particular item, and a number of times the particular item has been purchased by the retail customer within a predetermined time period.

4. The method of claim 1 , wherein outputting the basket prediction comprises providing the basket prediction to the retail website.

5. The method of claim 1 , wherein the identifier of the retail customer links the retail customer to item purchases based on matching of payment information to the retail customer and to the item purchases.

6. The method of claim 1 , further comprising obtaining promotion information associated with at least one item, and wherein the basket prediction includes one or more items associated with a promotion.

7. The method of claim 1 , wherein the basket prediction includes one or more items identified to be frequently purchased in combination with items previously purchased by the retail customer.

8. The method of claim 1 , further comprising obtaining feedback from a retail customer, wherein the feedback is to remove an item category from further basket prediction.

9. The method of claim 1 , wherein the carousel is configured to display one item within a plurality of item categories and presenting the carousel including at least one item comprises presenting only a highest-likelihood item to be purchased from each of the plurality of item categories.

10. A retail item recommendation system comprising: a recommendation modeling computing system hosting a recommendation Application Programming Interface (API) exposed to a retail website server, the recommendation API configured to receive information identifying a retail customer and output a basket prediction for that retail customer;
a recommendation modeling engine executing on the recommendation modeling computing system, the recommendation modeling engine configured to, when executed by a programmable circuit of the recommendation modeling computing system, perform:
identifying sales data associated with a retail customer based on historical sales being associated with an identifier of the retail customer, the sales data including recurring sales data representing sales of items purchased a plurality of times within a predefined time period;
classifying the sales data into item level data, wherein the item level data includes frequency and timing of the retail customer's historical purchase of items;
for each item, identifying an item category;
for each retail customer and each item category, building a time-series model of purchases;
determining, from the time-series model of purchases for the retail customer and item category, one or more model features;
generating a plurality of time-series models from the one or more model features for the retail customer, the plurality of time-series models including a date model generated using an autoregressive integrated moving average (ARIMA) model and a r

Assignees

• Target Brands Inc

Inventors

• Pande Amit
• Portnoy Jacob
• Siskind Ryan
• Copeland Brian

Classifications

• G06Q30/0631 Primary

  Recommending goods or services · CPC title

• G06Q30/0201 Primary

  Market modelling; Market analysis; Collecting market data · CPC title

• G06Q30/0224

  based on user history · CPC title

• G06Q30/0643

  graphically representing goods, e.g. 3D product representation · CPC title