Ce diaporama a bien été signalé.
Nous utilisons votre profil LinkedIn et vos données d’activité pour vous proposer des publicités personnalisées et pertinentes. Vous pouvez changer vos préférences de publicités à tout moment.

Tag Maps

4 684 vues

Publié le

Generating Summaries and Visualization for Large Collections of Geo-referenced Photographs

Publié dans : Voyages, Business

Tag Maps

  1. 1. Generating Summaries and Visualization for Large Collections of Geo-referenced Photographs Alexander Jaffe*, Mor Naaman *, Tamir Tassa † , Marc Davis $ *Yahoo! Research Berkeley † Open University of Israel $ Yahoo! Research
  2. 2. Attraction Map of Paris <ul><ul><li>Stanley Milgram, 1976. </li></ul></ul><ul><ul><li>Psychological Maps of Paris </li></ul></ul>
  3. 3. Attraction Map of London <ul><ul><li>Jaffe et al, 2006. </li></ul></ul>
  4. 4. Information Overload? <ul><ul><li>Flickr “geotagged” </li></ul></ul>
  5. 5. Overview <ul><li>Problem definition </li></ul><ul><li>Intuition for solution </li></ul><ul><li>Algorithm for summarization </li></ul><ul><li>Visualizing the dataset </li></ul><ul><li>Evaluation </li></ul><ul><li>Demo? </li></ul>
  6. 6. Problem Definition <ul><li>Dataset: </li></ul><ul><li>( photo_id , user_id, latitude, longitude) </li></ul><ul><li>( photo_id , tag ) </li></ul><ul><li>Result: </li></ul><ul><li>(photo_id, rank) </li></ul><ul><ul><li>Given all photos from a geographic region, find a “representative” summary set </li></ul></ul>
  7. 7. Issues to Tackle <ul><li>Noisy data </li></ul>Whatever, color, city, spectrum, santa barbara, california, usa, Lookatme, Herbert Bayer Chromatic Gate <ul><li>Photographer biases </li></ul><ul><ul><li>In locations </li></ul></ul><ul><ul><li>In Tags </li></ul></ul><ul><li>Wrong data </li></ul>
  8. 8. Intuition <ul><ul><li>More “activity” in a certain location indicates importance of that location </li></ul></ul><ul><ul><li>Tag that are unique to a certain location can suggest importance of that location </li></ul></ul>
  9. 9. (Very) Simple Example
  10. 10. Algorithm Overview <ul><li>Hierarchical Clustering of the location data </li></ul><ul><li>For each cluster, generate cluster score </li></ul><ul><li>Recursively generate ordering of all photos in each cluster, based on subcluster score and ordering </li></ul>
  11. 11. The Clustered Return of the (Very) Simple Example! 4, 6, 5 8,7 4,8,6,5,7 20 10
  12. 12. Generating a Summary <ul><li>A complete ranking is produced for all photos in the dataset </li></ul><ul><li>An n -photo summary is simply the first n photos in this ranking. </li></ul>
  13. 13. Generating Cluster Scores <ul><li>Main Factors: </li></ul><ul><ul><li>Number of photos </li></ul></ul><ul><ul><li>Relevance (bias) factors </li></ul></ul><ul><ul><li>“ Tag Distinguishability” </li></ul></ul><ul><ul><li>“ Photographer Distinguishability” </li></ul></ul>
  14. 14. Tag Distinguishability <ul><li>A measure of uniqueness of concepts represented in the cluster (“document”) </li></ul><ul><li>TF/IDF based </li></ul><ul><ul><li>Compute frequency of each tag (TF) </li></ul></ul><ul><ul><li>Compute (inverse) frequency of tag in the rest of the dataset (IDF) </li></ul></ul><ul><ul><li>Aggregate TF/IDF over all tags in cluster using L2 norm </li></ul></ul><ul><li>Or, if you like formulas: </li></ul>Read the damn paper!
  15. 15. Summary of San Francisco Golden Gate Bridge TransAmerica AT&T Baseball Park Golden Gate Twin Peaks Golden Gate Bay Bridge Ocean Beach Chinatown
  16. 16. Progress Bar (almost done) <ul><li>Problem definition </li></ul><ul><li>Intuition for solution </li></ul><ul><li>Algorithm for summarization </li></ul><ul><li>Visualizing the dataset </li></ul><ul><li>Evaluation </li></ul><ul><li>Demo? </li></ul>
  17. 17. Tag Maps <ul><li>Observation: </li></ul><ul><ul><li>The algorithm identifies “representative” locations </li></ul></ul><ul><ul><li>The algorithm identifies unique, important tags </li></ul></ul>Can be used to visualize the dataset!
  18. 18. Tag Maps
  19. 19. Tag Maps
  20. 20. Ok, how do we evaluate this? <ul><li>Direct human-evaluation of algorithmic results </li></ul><ul><ul><li>Evaluated Tag Maps with various weighting options </li></ul></ul><ul><ul><li>Compared summaries to 3 base conditions </li></ul></ul><ul><li>Compared chosen locations to top 15 locations selected by humans (Milgram-style) </li></ul>
  21. 21. Maybe we have time for a demo
  22. 22. Maybe we have time for Q’s <ul><li>http://zonetag.research.yahoo.com </li></ul><ul><li>(applied in prototype cameraphone app) </li></ul><ul><li>http://blog.yahooresearchberkeley.com </li></ul><ul><li>(more on this and other topics) </li></ul><ul><li>Become an intern, get involved: </li></ul><ul><li>Email me. </li></ul><ul><li>Mor Naaman </li></ul><ul><li>[email_address] </li></ul>

×