Generating recommended items in unfamiliar domain

Patent metadata

Field	Value
Publication number	US-9785954-B2
Application number	US-201113034849-A
Country	US
Kind code	B2
Filing date	Feb 25, 2011
Priority date	Feb 26, 2010
Publication date	Oct 10, 2017
Grant date	Oct 10, 2017

Abstract

A method of generating recommended items for a current user in an unfamiliar domain includes selecting a reference user in a reference domain different from the unfamiliar domain, where current user behavior in the reference domain and reference user behavior in the reference domain have a user similarity index which satisfies a user similarity index condition. The method further includes generating the recommended items in the unfamiliar domain for the current user according to history behavior data of the reference user in the unfamiliar domain.

First claim

That which is claimed is:

1. A method of generating recommended items for a current user in an unfamiliar domain, comprising:
selecting a plurality of candidate reference users sharing a user type with the current user in a reference domain, wherein the reference domain differs from the unfamiliar domain;
calculating a plurality of user similarity indices for the plurality of candidate reference users, comprising calculating a user similarity index for a candidate reference user with respect to a reference domain and the current user, wherein the user similarity index is based at least in part on common behaviors between the current user and the candidate reference user in the reference domain, and is calculated according to history behavior data of a plurality of demarcated users in the unfamiliar domain and the reference domain, wherein the plurality of demarcated users are similar users in two or more domains that are different than the current user;
selecting the candidate reference user as a reference user in the reference domain, responsive to the user similarity index of the candidate reference user satisfying a user similarity index condition;
determining a domain similarity index between the unfamiliar domain and the reference domain, wherein the domain similarity index is distinct from the user similarity index, and wherein the domain similarity index indicates a degree of similarity between the unfamiliar domain and the reference domain;
calculating a user weight for the reference user with respect to the current user, wherein the user weight is calculated by multiplying the user similarity index by the domain similarity index;
generating two or more candidate items in the unfamiliar domain for the current user according to history behavior data of other users, comprising at least the reference user, in the unfamiliar domain;
assigning an item weight to each of the two or more candidate items, wherein for each of the two or more candidate items, the item weight assigned is calculated by adding the user weights of the other users who are associated with the candidate item; and
selecting, as the recommended items, a subset of the candidate items whose item weights meet an item weight condition.

2. The method according to claim 1 , wherein the reference user in the reference domain has a number of behaviors common to the current user in the reference domain such that a common behavior number condition is satisfied.

3. The method according to claim 1 , wherein the reference user has a score for behaviors common to the current user in the reference domain, and wherein the score satisfies a common behavior score condition in the reference domain.

4. The method according to claim 1 , wherein determining the domain similarity index between the unfamiliar domain and the reference domain is at least partially based on history behavior data of a demarcated user in the unfamiliar domain and history behavior data of a demarcated user in the reference domain, wherein none of the unfamiliar domain and the reference domain is an unfamiliar domain for the demarcated user.

5. The method according to claim 1 , wherein calculating the user similarity index for the reference user with respect to the reference domain and the current user comprises:
determining a current purchase vector for the current user, wherein each element of the current purchase vector corresponds to a purchasable item, and wherein the value of each element in the current purchase vector indicates whether the current user has purchased the purchasable item corresponding to the element;
determining a reference purchase vector for the reference user, wherein each element of the reference purchase vector corresponds to a purchasable item, and wherein the value of each element in the reference purchase vector indicates whether the reference user has purchased the purchasable item corresponding to the element; and
calculating an irrelevancy index based on a sum of the squares of the differences between the elements of the current purchase vector and the reference purchase vector;
wherein the user similarity index is based at least in part on the irrelevancy index.

6. An apparatus for generating recommended items for a current user in an unfamiliar domain, comprising:
a reference user determining module configured to:
select a plurality of candidate reference users sharing a user type with the current user in a reference domain, wherein the reference domain differs from the unfamiliar domain;
calculate a plurality of user similarity indices for the plurality of candidate reference users, comprising calculating a user similarity index for a candidate reference user with respect to a reference domain and the current user, wherein the user similarity index is based at least in part on common behaviors between the current user and the candidate reference user in the reference domain, and is calculated according to history behavior data of a plurality of demarcated users in the unfamiliar domain and the reference domain, wherein the plurality of demarcated users are similar users in two or more domains that are different than the current user; and
select the candidate reference user as a reference user in the reference domain, responsive to the user similarity index of the candidate reference user satisfying a user similarity index condition;
a domain similarity determining module configured to determine a domain similarity index between the unfamiliar domain and the reference domain, wherein the domain similarity index is distinct from the user similarity index, and wherein the domain similarity index indicates a degree of similarity between the unfamiliar domain and the reference domain; and
one or more computer processors configured to:
calculate a user weight for the reference user with respect to the current user, wherein the user weight is calculated by multiplying the user similarity index by the domain similarity index;
generate two or more candidate items in the unfamiliar domain for the current user according to history behavior data of other users, comprising at least the reference user, in the unfamiliar domain;
assign an item weight to each of the two or more candidate items, wherein for each of the two or more candidate items, the item weight assigned is calculated by adding the user weights of the other users who are associated with the candidate item; and
select, as the recommended items, a subset of the candidate items whose item weights meet an item weight condition.

7. The apparatus according to claim 6 , wherein the reference user has a number of behaviors common to the current user that satisfies a common behavior number condition in the reference domain.

8. The apparatus according to claim 6 , wherein the reference user has scores for behaviors common to the current user that satisfy a common behavior score condition in the reference domain.

9. The apparatus according to claim 6 , further comprising:
a demarcating module configured to determine the domain similarity index between the unfamiliar domain and the reference domain according to history behavior data of a demarcated user in the unfamiliar domain and the reference domain, wherein none of the unfamiliar domain and the reference domain is an unfamiliar domain for the demarcated user.

10. The apparatus according to claim 6 , wherein the reference user determining module is further configured to:
determine a current purchase vector for the current user, wherein each element of the current purchase vector corresponds to a purchasable item, and wherein the value of each element in the current purchase vector indicates whether the current user has purchased the purchasable it

Assignees

• Wu Xian
• Yuan Quan
• Zhang Xia Tian
• Zhao Shiwan
• IBM

Inventors

• Wu Xian
• Yuan Quan
• Zhang Xia Tian
• Zhao Shiwan

Classifications

• G06F16/335

  Filtering based on additional data, e.g. user or group profiles (filtering in web context G06F16/9535, G06F16/9536) · CPC title

• G06Q30/02 Primary

  Marketing; Price estimation or determination; Fundraising · CPC title

• G06F17/30699

  Physics · mapped topic