Recommendation generation for infrequently accessed items

Patent metadata

Field	Value
Publication number	US-9959563-B1
Application number	US-201314135176-A
Country	US
Kind code	B1
Filing date	Dec 19, 2013
Priority date	Dec 19, 2013
Publication date	May 1, 2018
Grant date	May 1, 2018

Abstract

Systems and methods are disclosed for generating recommendation rules based on the attributes of items that are purchased together at a threshold rate. The attributes of the items may be extracted from item-detail content associated with the items. Using a count of the frequency with which pairs of items include pairs of attributes, a recommendation rule can be created that recommends items with particular attributes to users who access other items with particular attributes. Further, using the recommendation rules, items may be selected for recommendation to users who access an item that lacks historical access data from which to generate recommendations solving the “cold-start” problem. Moreover, negative rules may be generated based on historical access data and attributes of items purchased and/or not purchased together at a threshold rate to prevent the recommendation of particular items to users who access items associated with the negative rules.

First claim

What is claimed is:

1. A computer-implemented method of determining an item to recommend, the computer-implemented method comprising:
detecting a behavior-based association between a first item facet and a second item facet based on selections by users of catalog items in an electronic catalog, wherein a first plurality of the catalog items have the first item facet and a second plurality of the catalog items have the second item facet the first item facet comprising a first attribute shared among a first plurality of items, and the second item facet comprising a second attribute shared among a second plurality of items, the second attribute differing from the first attribute;
identifying a first catalog item having the first item facet;
determining a plurality of recommendation rules for the first catalog item, wherein performance of different recommendation rules of the plurality of recommendation rules results in identification of different sets of items for recommendation to users who access the first catalog item;
scoring each of the plurality of recommendation rules to obtain a score for each recommendation rule, the score generated based at least in part on a ratio of posterior odds to prior odds determined from historical item-access data associated with the recommendation rule;
classifying the plurality of recommendation rules based at least in part on a determination of whether the score associated with each of the recommendation rules satisfies a score threshold to obtain a first set of recommendation rules that satisfy the score threshold and a second set of recommendation rules that do not satisfy the score threshold, wherein the first set of recommendation rules comprise positive recommendation rules and the second set of recommendation rules comprise negative recommendation rules;
receiving additional item-access data associated with at least one recommendation rule from the second set of recommendation rules;
rescoring the at least one recommendation rule based on the additional item-access data to obtain an updated score;
reclassifying the at least one recommendation rule based at least in part on a determination that the updated score satisfies the score threshold, wherein reclassifying the at least one recommendation rule comprises removing the at least one recommendation rule from the second set of recommendation rules and adding the at least one recommendation rule to the first set of recommendation rules;
selecting a recommendation rule from the first set of recommendation rules based at least in part on the score associated with the recommendation rule;
selecting, using the selected recommendation rule and based at least partly on the behavior-based association between the first and second item facets, a second catalog item to recommend to users who select the first catalog item, said second catalog item having the second facet; and
outputting for recommendation the second catalog item to a user who selected the first catalog item,
wherein the computer-implemented method is performed programmatically by a computing system comprising one or more computing devices.

2. The computer-implemented method of claim 1 , wherein the selections comprise purchases.

3. The computer-implemented method of claim 1 , wherein selecting the second catalog item comprises selecting a top ranked item from the second plurality of the catalog items that have the second item facet.

4. The computer-implemented method of claim 1 , wherein selecting the second catalog item comprises relying on the behavior-based association between the first and second item facet in response to determining that a recommendation based solely on item-to-item behavioral associations would be unreliable.

5. The computer-implemented method of claim 1 , wherein the first item facet comprises an item category identifier in combination with the first attribute.

6. The computer-implemented method of claim 1 , wherein the first and second item facets correspond to different respective item categories.

7. The computer-implemented method of claim 1 , wherein detecting the behavior-based association comprises comparing (1) a measure of how frequently a catalog item having the first item facet is purchased together with a catalog item having the second item facet, to (2) a measure of how frequently a catalog item having the first item facet is purchased together with a catalog item not having the second item facet.

8. The computer-implemented method of claim 1 , further comprising supplementing an electronic catalog page corresponding to the first catalog item with a recommendation of the second catalog item.

9. The computer-implemented method of claim 8 , wherein the recommendation comprises an indication that users who purchase items having the first item facet also purchase items having the second item facet.

10. A computer-implemented method of determining a set of items to recommend in association with behaviorally-deficient items, the computer-implemented method comprising:
as implemented by a computing system including one or more hardware processors,
identifying a first item for which item-level association data is deficient, wherein the item-level association data identifies items that are related to each other based on a determined relationship between a first item facet and a second item facet and based on selections by users of catalog items in an electronic catalog, wherein a first plurality of the catalog items have the first item facet and a second plurality of the catalog items have the second item facet, the first item facet comprising a first attribute shared among a first plurality of items, and the second item facet comprising a second attribute shared among a second plurality of items, the second attribute differing from the first attribute, and wherein the first item is determined to be associated with a set of items based on a history of transactions that include the first item and an item from the set of items;
extracting an attribute for the first item;
determining an item category for the first item;
accessing a set of recommendation rules associated with the item category, the set of recommendation rules generated based on attribute pairs comprising the first attribute and the second attribute and based on a determination that users who access an item having the first attribute access an item having the second attribute at a threshold rate, wherein the first attribute is associated with at least some items of the item category and the second attribute is associated with at least some items of another item category, wherein performance of different recommendation rules of the set of recommendation rules results in identification of different sets of items for recommendation to users who access the first item;
scoring each of the set of recommendation rules to obtain a score for each recommendation rule, the score generated based at least in part on a ratio of posterior odds to prior odds determined from historical item-access data associated with the recommendation rule;
classifying at least one recommendation rule from the set of recommendation rules as a positive recommendation rule based at least in part on a determination that the score associated with the at least one recommendation rule satisfies a score threshold;
receiving additional item-access data associated with the at least one recommendation rule from the set of recommendation rules;
rescoring the at least one recommendation rule based on the additional item-access data to obtain an updated score;
reclassifying the at least one recommendation rule as a negative recommendation rule based at least in part on a determination that the updated score does not s

Assignees

• Amazon Tech Inc

Inventors

• Wu Jianhui
• Patnaik Debprakash
• Wang Rui

Classifications

• G06Q30/0631 Primary

  Recommending goods or services · CPC title