BEHAVIOR BASED METHOD AND SYSTEM FOR FILTERING OUT UNFAIR RATINGS FOR TRUST MODELS

• US 20090327181A1
• Filed: 06/30/2009
• Published: 12/31/2009
• Est. Priority Date: 06/30/2008
• Status: Abandoned Application

• Current Assignee: Industry-Academic Cooperation Foundation Yonsei University
• Original Assignee: Industry-Academic Cooperation Foundation Yonsei University
• Inventors: Yuan, Wei Wei, Lee, Sung-Young, Lee, Young-Koo
• Application Number:
  US12/494,446
• Publication Number:
  US 20090327181A1
• Time in Patent Office: Days
• US Class Current: 706/20
• CPC Class Codes: G06Q 10/0637 Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0278 Product appraisal

Abstract

Disclosed is a behavior-based method which uses each rater'"'"'s rating behaviors as the criterion to judge unfair ratings. A behavior refers to the action that a rater gives certain rating under specific context. The behavior-based method regards the rating given by a rater with abnormal behavior as an unfair rating, where abnormal behavior is recognized by comparing a rater'"'"'s current behavior with his behavior history.

Claims

1. A method for filtering out unfair ratings based on behaviors, comprising:

receiving ratings and contexts under which the ratings were given;

classifying raters into fair raters with fair behavior and doubtful raters with doubtful behavior using the ratings and contexts, ratings from the fair raters being fair ratings and ratings from the doubtful raters being doubtful ratings;

calculating a truster'"'"'s final trust decision on the ratee using ratings given by the fair raters; and

regarding each of the doubtful raters as an unfair rater if the received rating is different from the truster'"'"'s final trust decision, and otherwise as a retrain rater, and regarding the rating from the unfair rater as an unfair rating.

2. The method of claim 1, wherein the classifying includes:

calculating an expected rating for each of the raters who gave the ratings for the context based on its judging rule;

comparing the expected rating with the received rating for each rater; and

regarding each of the raters as a doubtful rater if the expected rating is different from the received rating, and otherwise as a fair rater and regarding the rating from the fair rater as a fair rating and the rating from the doubtful rater as a doubtful rating.

3. The method of claim 2, wherein the judging rule is learned using incremental learning neural network.

4. The method of claim 3, further comprising retraining the judging rule of the retrain rater by inputting the received rating of the retrain rater into the incremental learning neural network.

5. The method of claim 4, wherein Cascade-Correlation architecture is used for the incremental learning neural network.

6. A system for filtering out unfair ratings based on behaviors, comprising:

means for receiving ratings and contexts under which the ratings were given;

means for classifying raters into fair raters with fair behavior and doubtful raters with doubtful behavior using the ratings and contexts, ratings from the fair raters being fair ratings and ratings from the doubtful raters being doubtful ratings;

means for calculating a truster'"'"'s final trust decision on the ratee using ratings given by the fair raters; and

means for regarding each of the doubtful raters as an unfair rater if the received rating is different from the truster'"'"'s final trust decision, and otherwise as a retrain rater, and regarding the rating from the unfair rater as an unfair rating.

7. The system of claim 6, wherein the means for classifying raters includes:

means for calculating an expected rating for each of the raters who gave the ratings for the context based on its judging rule;

means for comparing the expected rating with the received rating for each rater; and

means for regarding each of the raters as a doubtful rater if the expected rating is different from the received rating, and otherwise as a fair rater and regarding the rating from the fair rater as a fair rating and the rating from the doubtful rater as a doubtful rating.

8. The method of claim 7, wherein the judging rule is learned using incremental learning neural network.

9. The method of claim 8, further comprising means for retraining the judging rule of the retrain rater by inputting the received rating of the retrain rater into the incremental learning neural network.

10. The method of claim 9, wherein Cascade-Correlation architecture is used for the incremental learning neural network.