US 2017/0287023 A1Published application

Blacklisting Based on Image Feature Analysis and Collaborative Filtering

AssigneeCriteo SA

InventorsOlivier Koch, Pierre-Emmanuel Mazaré, Romain Lerallut, Dan Teodosiu

Life of this patent

Estimating the term…

Loading patent…

Abstract

Described are methods, systems, and apparatus for recommending catalogue items for blacklisting. For each of a plurality of catalog items: image features are extracted; image features are associated with the catalog item; and user purchase events, user view events, a textual description, and categories are associated with the catalog item by the recommendation system. An identification of a first blacklisted catalog item is received. A catalog item is identified by the recommendation system based on i) a similarity between the image features associated with the first blacklisted catalog item and the image features associated with the catalog item, and ii) a correspondence between at least one of the user purchase events, the user view events, the textual description, or the categories associated with the blacklisted catalog item and the user purchase events, the user view events, the textual description, or the categories associated with the catalog item.

Claims14

Dependent claims

A computer-implemented method of recommending one or more catalogue items for blacklisting, the method comprising:

a. for each catalog item of a plurality of catalog items:
i. extracting, by the recommendation system, from at least one image associated with the catalog item, one or more catalog item image features;
ii. associating, by the recommendation system, with the catalog item, the one or more catalog item image features;
iii. associating, by the recommendation system, with the catalog item, at least one of user purchase events for a plurality of users, user view events for a plurality of users, a textual description of the catalog item, and one or more categories for the catalog item;
b. receiving, by the recommendation system, an identification of a first blacklisted catalog item, the first blacklisted catalog item being a catalog item of the plurality of catalog items;
c. identifying, by the recommendation system, a first catalog item of the plurality of catalog items as a first candidate blacklisted catalog item based on i) a similarity between the one or more catalog item image features associated with the first blacklisted catalog item and the one or more image features associated with the first catalog item, and ii) a correspondence between at least one of the user purchase events, the user view events, the textual description, or the one or more categories associated with the blacklisted catalog item and at least one of the user purchase events, the user view events, the textual description, or the one or more categories associated with the first catalog item.

3 more elements

A computer-implemented method of recommending one or more catalogue items for blacklisting, the method comprising:

a. storing, by the recommendation system, a plurality of non-blacklisted catalog items, each catalog item of the plurality of non-blacklisted catalog items associated with one or more catalog item image features extracted from at least one image associated with the catalog item, and associated with at least one of user purchase events for a plurality of users, user view events for a plurality of users, a textual description of the catalog item, and one or more categories for the catalog item;
b. storing, by the recommendation system, a plurality of blacklisted catalog items, each catalog item of the plurality of blacklisted catalog items associated with one or more catalog item image features extracted from at least one image associated with the catalog item, and associated with at least one of user purchase events for a plurality of users, user view events for a plurality of users, a textual description of the catalog item, and one or more categories for the catalog item;
c. identifying, by the recommendation system, a plurality of candidate blacklisted catalog items from the plurality of non-blacklisted catalog items based on i) a similarity between the one or more catalog item image features associated with at least one of the plurality of blacklisted catalog items and the one or more image features associated with at least one of the plurality of candidate blacklisted catalog items, and ii) a correspondence between at least one of the user purchase events, the user view events, the textual description of the catalog item, or the one or more categories associated with the at least one of the plurality of blacklisted catalog items, and at least one of the user purchase events, the user view events, the textual description of the catalog item, or the one or more categories associated with the at least one of the plurality of candidate blacklisted catalog items.

Dependencies are inferred from the claim text.

Facts

Application: 15/087,714Filed 31 Mar 2016
Published: 5 Oct 2017
Term ends: Not estimatedThe source records don’t support an estimate.

Classification7 codes

G06Q 30/0601
Buying, selling or leasing transactions: electronic shopping [e-shopping]
G06F 18/2178
Validation; Performance evaluation; Active pattern learning techniques: based on feedback of a supervisor
G06V 10/7784
Active pattern-learning, e.g. online learning of image or video features: based on feedback from supervisors
G06V 20/10
Scenes; Scene-specific elements: terrestrial scenes

Family

US 10,290,032 B2Related patent
US 2017/0287023 A1This patent

From the US filing history. Foreign family members aren’t listed.

Timeline

Citations

Open elsewhere

Building an agent or pulling many patents? Use the API or MCP server instead of scraping pages. Get an API key