MULTI-LEVEL REPUTATION BASED RECOMMENDATION SYSTEM AND METHOD

• US 20090177691A1
• Filed: 01/03/2008
• Published: 07/09/2009
• Est. Priority Date: 01/03/2008
• Status: Active Grant

• Current Assignee: Gary Manfredi, Jakob Darger, Kim Scheibel
• Original Assignee: Gary Manfredi, Jakob Darger, Kim Scheibel
• Inventors: Darger, Jakob, Manfredi, Gary, Scheibel, Kim
• Granted Patent: US 7,734,609 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/958 Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking

Abstract

Multi-level reputation based recommendation system and method focused on users as opposed to products or web sites. Establishes reputations of recommenders that may extend beyond a given user'"'"'s first hand trust of these recommenders. Trust formed between a second and third user based on similar valuations that may not be shared between the first user and the third user may be utilized when recommending something useful for the first user. Reputations are subjective to each user and change over time. Provides recommendations to connect users with blogs, videos, other users, music, books, web sites, reviews, products, vacations spots, hotels, cities, events, activities and ad units for example. Does not rely on the number of static information connections, i.e., links to a given web site, but rather utilize dynamic information as it spreads through the interaction of users. May be utilized in any application or domain where users attribute value to things or actions.

Claims

1. A reputation based recommendation system comprising:

a computer memory;

a computer processing unit;

a computer comprising said computer memory and said computer processing unit wherein said computer is configured to;

obtain a first rating from a first user on a first item;

obtain a second rating from a second user on said first item;

obtain a third rating from said second user on a second item;

obtain a fourth rating from a third user on said second item;

assign said first rating and said second rating on said first item to item attributes associated with said first item and assign said third rating and said fourth rating to item attributes associated with said second item;

calculate a first consideration of ratings obtained from said first user on a first item attribute selected from attributes associated with said first item;

calculate a second consideration of ratings obtained from said second user on said first item attribute selected from attributes associated with said first item;

calculate a third consideration of ratings obtained from said second user on a second item attribute selected from attributes associated with said second item;

calculate a fourth consideration of ratings obtained from said third user on said second item attribute selected from attributes associated with said second item;

calculate a first agreement between said first user and said second user via said first consideration and said second consideration wherein similar considerations provide a higher level of agreement than dissimilar considerations;

calculate a second agreement between said second user and said third user via said third consideration and said fourth consideration wherein similar considerations provide said higher level of agreement than said dissimilar considerations;

calculate agreement sums for all agreements between said first user and said second user and said second user and said third user;

calculate an opinion of said first user with respect to said second user and said second user with respect to said third user based on said agreement sums;

calculate a reputation of said third user with respect to said first user based on opinions between said first user and said second user and between said second user and said third user based on network path analysis where users are vertices and opinions are edges wherein said reputation is formed via summation of values of acyclic paths which connect said first user and said third user, and where each acyclic path value is calculated as a product of weights of at least one edge or opinion that makes up said acyclic path and where each hop is adjusted by a normalization factor that represents a distance from said first user; and

,

calculate at least one prediction for said first user on an arbitrary item by calculating a reputation weighted average of a plurality of ratings on said arbitrary item or attribute associated with said item by users who have a reputation with respect to said first user.

2. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said first rating via a rating confidence factor wherein said rating confidence factor is a function of time and wherein said rating confidence factor decreases in value with time.

3. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said first rating via a rating confidence factor wherein said rating confidence factor does not fall below a level when a number of votes is below a threshold.

4. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said first consideration via normalization of each confidence by a largest confidence in an attribute class.

5. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said first agreement through a confidence weighting.

6. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said agreement sums by a attribute class weight factor.

7. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

add said second user to a list of favorite users for said first user.

8. The reputation based recommendation system of claim 1 wherein said computer is further configured to:

modify said prediction via a sum of reputation weighted consideration confidences for a given attribute.

9. A reputation based recommendation method comprising:

obtaining a first rating from a first user on a first item;

obtaining a second rating from a second user on said first item;

obtaining a third rating from said second user on a second item;

obtaining a fourth rating from a third user on said second item;

assigning said first rating and said second rating on said first item to item attributes associated with said first item and assigning said third rating and said fourth rating to item attributes associated with said second item;

calculating a first consideration of ratings obtained from said first user on a first item attribute selected from attributes associated with said first item;

calculating a second consideration of ratings obtained from said second user on said first item attribute selected from attributes associated with said first item;

calculating a third consideration of ratings obtained from said second user on a second item attribute selected from attributes associated with said second item;

calculating a fourth consideration of ratings obtained from said third user on said second item attribute selected from attributes associated with said second item;

calculating a first agreement between said first user and said second user via said first consideration and said second consideration wherein similar considerations provide a higher level of agreement than dissimilar considerations;

calculating a second agreement between said second user and said third user via said third consideration and said fourth consideration wherein similar considerations provide said higher level of agreement than said dissimilar considerations;

calculating agreement sums for a plurality of agreements between said first user and said second user and said second user and said third user;

calculating an opinion of said first user with respect to said second user and said second user with respect to said third user based on said agreement sums;

calculating a reputation of said third user with respect to said first user based on opinions between said first user and said second user and between said second user and said third user based on network path analysis where users are vertices and opinions are edges wherein said reputation is formed via summation of values of acyclic paths which connect said first user and said third user, and where each acyclic path value is calculated as a product of weights of at least one edge or opinion that makes up said acyclic path and where each hop is adjusted by a normalization factor that represents a distance from said first user; and

,

calculating at least one prediction for said first user on an arbitrary item by calculating a reputation weighted average of a plurality of ratings on said arbitrary item or attribute associated with said item by users who have a reputation with respect to said first user.

10. The reputation based recommendation method of claim 9 further comprising:

modifying said first rating via a rating confidence factor wherein said rating confidence factor is a function of time and wherein said rating confidence factor decreases in value with time.

11. The reputation based recommendation method of claim 9 further comprising:

modifying said first rating via a rating confidence factor wherein said rating confidence factor does not fall below a level when a number of votes is below a threshold.

12. The reputation based recommendation method of claim 9 further comprising:

modifying said first consideration via normalization of each confidence by a largest confidence in an attribute class.

13. The reputation based recommendation method of claim 9 further comprising:

modifying said first agreement through a confidence weighting.

14. The reputation based recommendation method of claim 9 further comprising:

modifying said agreement sums by a attribute class weight factor.

15. The reputation based recommendation method of claim 9 further comprising:

adding said second user to a list of favorite users for said first user.

16. The reputation based recommendation method of claim 9 further comprising:

modifying said prediction via a sum of reputation weighted consideration confidences for a given attribute.