User-powered recommendation system

Patent metadata

Field	Value
Publication number	US-10311367-B2
Application number	US-201614988997-A
Country	US
Kind code	B2
Filing date	Jan 6, 2016
Priority date	Nov 17, 2008
Publication date	Jun 4, 2019
Grant date	Jun 4, 2019

Abstract

Recommendation systems are widely used in Internet applications. In current recommendation systems, users only play a passive role and have limited control over the recommendation generation process. As a result, there is often considerable mismatch between the recommendations made by these systems and the actual user interests, which are fine-grained and constantly evolving. With a user-powered distributed recommendation architecture, individual users can flexibly define fine-grained communities of interest in a declarative fashion and obtain recommendations accurately tailored to their interests by aggregating opinions of users in such communities. By combining a progressive sampling technique with data perturbation methods, the recommendation system is both scalable and privacy-preserving.

First claim

The invention claimed is:

1. A method comprising:
receiving a user defined declarative community definition comprising a predicate on user attributes;
selecting a plurality of members, based on the user defined declarative community definition, split into a plurality of disjoint sampling groups, each particular disjoint sampling group in the plurality of disjoint sampling groups having a particular size, the particular size decreases exponentially from a first size to a second size, and the plurality of members selected from the disjoint sampling group with the smallest size such that an accuracy level of a plurality of perturbed rating data is greater than or equal to a user-defined accuracy level;
for each member of the plurality of members, retrieving, from a recommendation database storing a plurality of data structures, a particular rating associated with each member based on the user-defined recommendation request to generate a plurality of ratings, each of the plurality of data structures comprising a sum over a random subset of original data;
generating a plurality of perturbed ratings based on the plurality of ratings such that there is a predetermined probability that each rating of the plurality of ratings is different from its respective perturbed rating of the plurality of perturbed ratings;
aggregating the plurality of perturbed ratings to generate an aggregated perturbed rating; and
generating a recommendation based on the aggregated perturbed rating.

2. The method of claim 1 , further comprising:
receiving a user-defined item constraint,
wherein the generating the recommendation is further based on the user-defined item constraint.

3. The method of claim 1 , further comprising:
receiving a user defined recommendation request issued by a user as a user query, the recommendation generated in response to receiving the user defined recommendation request.

4. An apparatus comprising:
a processor; and
a memory to store computer program instructions, the computer program instructions when executed on the processor cause the processor to perform operations comprising:
receiving a user defined declarative community definition comprising a predicate on user attributes;
selecting a plurality of members, based on the user defined declarative community definition, split into a plurality of disjoint sampling groups, each particular disjoint sampling group in the plurality of disjoint sampling groups having a particular size, the particular size decreases exponentially from a first size to a second size, and the plurality of members selected from the disjoint sampling group with the smallest size such that an accuracy level of a plurality of perturbed rating data is greater than or equal to a user-defined accuracy level;
for each member of the plurality of members, retrieving, from a recommendation database storing a plurality of data structures, a particular rating associated with each member based on the user-defined recommendation request to generate a plurality of ratings, each of the plurality of data structures comprising a sum over a random subset of original data;
generating a plurality of perturbed ratings based on the plurality of ratings such that there is a predetermined probability that each rating of the plurality of ratings is different from its respective perturbed rating of the plurality of perturbed ratings;
aggregating the plurality of perturbed ratings to generate an aggregated perturbed rating; and
generating a recommendation based on the aggregated perturbed rating.

5. The apparatus of claim 4 , the operations further comprising:
receiving a user-defined item constraint,
wherein the generating the recommendation is further based on the user-defined item constraint.

6. The apparatus of claim 4 , the operations further comprising:
receiving a user defined recommendation request issued by a user as a user query, the recommendation generated in response to receiving the user defined recommendation request.

7. A non-transitory computer readable medium storing computer program instructions, which, when executed on a processor, cause the processor to perform operations comprising:
receiving a user defined declarative community definition comprising a predicate on user attributes;
selecting a plurality of members, based on the user defined declarative community definition, split into a plurality of disjoint sampling groups, each particular disjoint sampling group in the plurality of disjoint sampling groups having a particular size, the particular size decreases exponentially from a first size to a second size, and the plurality of members selected from the disjoint sampling group with the smallest size such that an accuracy level of a plurality of perturbed rating data is greater than or equal to a user-defined accuracy level;
for each member of the plurality of members, retrieving, from a recommendation database storing a plurality of data structures, a particular rating associated with each member based on the user-defined recommendation request to generate a plurality of ratings, each of the plurality of data structures comprising a sum over a random subset of original data;
generating a plurality of perturbed ratings based on the plurality of ratings such that there is a predetermined probability that each rating of the plurality of ratings is different from its respective perturbed rating of the plurality of perturbed ratings;
aggregating the plurality of perturbed ratings to generate an aggregated perturbed rating; and
generating a recommendation based on the aggregated perturbed rating.

8. The non-transitory computer readable medium of claim 7 , further comprising:
receiving a user-defined item constraint,
wherein the generating the recommendation is further based on the user-defined item constraint.

Assignees

• At & T Ip I Lp
• Univ Texas

Inventors

• Ramakrishnan Kadangode K
• Srivastava Divesh
• Cho Tae Won
• Zhang Yin

Classifications

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• H04N21/4661

  Deriving a combined profile for a plurality of end-users of the same client, e.g. for family members within a home (user profiles in network data switching protocols H04L67/306) · CPC title

• G11B27/105 Primary

  of operating discs · CPC title

• G06F16/48

  Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually · CPC title

• G06F16/244

  Grouping and aggregation · CPC title