Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Ranking System based on Social Network of Domain Experts Mateusz Marmołowski May 28, 2008
Motivation <ul><li>Rating / Ranking / Voting </li></ul><ul><li>Insufficient rating algorithms </li></ul><ul><li>Real-life ...
Our goals <ul><li>New, real-life rating algorithm </li></ul><ul><li>Fully automated and efficient approach </li></ul><ul><...
Rating Algorithm - agenda <ul><li>Rating Calculation </li></ul><ul><ul><li>Friendship influence </li></ul></ul><ul><ul><li...
Rating Algorithm
Rating Calculation
RC – Friendship influence Range: [50 – 100%]
RC – Domain knowledge <ul><li>Domain expertise </li></ul><ul><li>P arameter  α  – for amateurs </li></ul><ul><li>Inaccurat...
RC – Multiple domains <ul><li>Resource with multiple domains assigned </li></ul><ul><li>Solution -  geometric series </li>...
RC – User Weight <ul><li>Each rate is being weighted with UW </li></ul><ul><ul><li>Frienship with the author </li></ul></u...
RC – Weighted Rate <ul><li>Weighted Rate for resource </li></ul><ul><li>Rate given by user  </li></ul><ul><li>User Weight ...
RC – Bayesian Weighted Rate - The average number of votes of all resources - The average rating of all resources’ ratings ...
Expertise Calculation
EC – Weighted Expertise Value <ul><li>Separately calculated for each domain and user </li></ul><ul><li>Based on: </li></ul...
EC – Final expertise calculation <ul><li>EXP value is used for User Weight calculation </li></ul><ul><li>EXP strongly depe...
Expertise propagation
Expertise propagation – step 1
Expertise propagation – step 2
Expertise propagation – step 3
Summary <ul><li>Implementation in progress </li></ul><ul><li>Integration with MarcOnt Portal </li></ul><ul><li>Masters The...
Upcoming SlideShare
Loading in …5
×

Real-life Rating Algorithm

4,064 views

Published on

Presentation of the Rating Algorithm based on Social Network of Domain Experts

Published in: Technology, Health & Medicine
  • Be the first to comment

Real-life Rating Algorithm

  1. 1. Ranking System based on Social Network of Domain Experts Mateusz Marmołowski May 28, 2008
  2. 2. Motivation <ul><li>Rating / Ranking / Voting </li></ul><ul><li>Insufficient rating algorithms </li></ul><ul><li>Real-life situations, human behaviours </li></ul><ul><ul><li>Frienship factor </li></ul></ul><ul><ul><li>Domain experts </li></ul></ul><ul><li>Masters Thesis ;-) </li></ul>
  3. 3. Our goals <ul><li>New, real-life rating algorithm </li></ul><ul><li>Fully automated and efficient approach </li></ul><ul><li>Friendship consideration – social network </li></ul><ul><li>Expertise collection </li></ul><ul><li>REST SOA for simple access </li></ul>
  4. 4. Rating Algorithm - agenda <ul><li>Rating Calculation </li></ul><ul><ul><li>Friendship influence </li></ul></ul><ul><ul><li>Domain knowledge </li></ul></ul><ul><ul><li>Multiple domains </li></ul></ul><ul><ul><li>User Weight </li></ul></ul><ul><ul><li>Weighted Rate </li></ul></ul><ul><ul><li>Bayesian Weighted Rate </li></ul></ul><ul><li>Expertise Calculation </li></ul><ul><ul><li>Weighted Expertise Value </li></ul></ul><ul><ul><li>Activity factor </li></ul></ul><ul><ul><li>Final expertise calculation </li></ul></ul><ul><ul><li>Expertise propagation process </li></ul></ul>
  5. 5. Rating Algorithm
  6. 6. Rating Calculation
  7. 7. RC – Friendship influence Range: [50 – 100%]
  8. 8. RC – Domain knowledge <ul><li>Domain expertise </li></ul><ul><li>P arameter α – for amateurs </li></ul><ul><li>Inaccurate domain adjustment </li></ul><ul><ul><li>Generalization and particularization </li></ul></ul><ul><ul><li>Expertise propagation </li></ul></ul>
  9. 9. RC – Multiple domains <ul><li>Resource with multiple domains assigned </li></ul><ul><li>Solution - geometric series </li></ul><ul><li>MD factor: </li></ul>39% 5 47% 4 58% 3 75% 2 100% 1 MD(d) d
  10. 10. RC – User Weight <ul><li>Each rate is being weighted with UW </li></ul><ul><ul><li>Frienship with the author </li></ul></ul><ul><ul><li>Expertise in resource’s domains </li></ul></ul><ul><li>User Weight for a specific resource: </li></ul>
  11. 11. RC – Weighted Rate <ul><li>Weighted Rate for resource </li></ul><ul><li>Rate given by user </li></ul><ul><li>User Weight </li></ul><ul><li>The author of </li></ul><ul><li>The number of rates for a resource </li></ul>
  12. 12. RC – Bayesian Weighted Rate - The average number of votes of all resources - The average rating of all resources’ ratings - The number of votes for this resource - The rating of this resource Final formula:
  13. 13. Expertise Calculation
  14. 14. EC – Weighted Expertise Value <ul><li>Separately calculated for each domain and user </li></ul><ul><li>Based on: </li></ul><ul><ul><li>BWR for each resource </li></ul></ul><ul><ul><li>Number of rates for that resource </li></ul></ul><ul><ul><li>Mutiple Domain factor </li></ul></ul><ul><li>Rating scale normalization </li></ul>
  15. 15. EC – Final expertise calculation <ul><li>EXP value is used for User Weight calculation </li></ul><ul><li>EXP strongly depends on WEV </li></ul><ul><li>WEV recalculation in case of new rates </li></ul><ul><li>Parameter α for amateurs (no published resources) </li></ul>
  16. 16. Expertise propagation
  17. 17. Expertise propagation – step 1
  18. 18. Expertise propagation – step 2
  19. 19. Expertise propagation – step 3
  20. 20. Summary <ul><li>Implementation in progress </li></ul><ul><li>Integration with MarcOnt Portal </li></ul><ul><li>Masters Thesis </li></ul><ul><li>Publication </li></ul><ul><ul><li>DERILion Deliverable D1.6.3 - Real-life Rating Algorithm </li></ul></ul><ul><li>Thank you. </li></ul>

×