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.

World Explorer (JCDL 2007 Best Paper)

1,688 views

Published on

Slides from my presentation at JCDL 2007.

The paper was titled "World Explorer: Visualizing Aggregate Data from Unstructured Text in Geo-Referenced Collections" and won the Vannevar Bush Best Paper award. You can read the full paper at http://www.rahulnair.net/files/JCDL07-ahern-WorldExplorer.pdf and also see a demo at http://tagmaps.research.yahoo.com/worldexplorer.php

Published in: Technology, Education
  • Be the first to comment

World Explorer (JCDL 2007 Best Paper)

  1. 1. World Explorer: Visualizing Aggregate Data from Unstructured Text in Geo-Referenced Collections Shane Ahern, Mor Naaman, Rahul Nair* & Jeannie Yang Yahoo! Research Berkeley
  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 Paris <ul><ul><li>Y!RB, 2007. </li></ul></ul>
  4. 4. Flickr “geotagged” <ul><ul><li>20+ million images </li></ul></ul>Can we do better?
  5. 5. Location-driven Modeling <ul><li>Derive meaningful data about map regions </li></ul><ul><li>E.g., representative tags, photos </li></ul>
  6. 6. Data Description
  7. 7. Issues <ul><li>Sparse data set </li></ul><ul><li>Photographer bias </li></ul><ul><ul><li>In location </li></ul></ul><ul><ul><li>In tags </li></ul></ul><ul><li>Incorrect data </li></ul>
  8. 8. Heuristics <ul><li>Number of photographs denotes the “importance” of a location </li></ul><ul><li>Users will use a common subset of tags to describe objects/locations </li></ul><ul><li>Concentrated tag usage indicates descriptiveness </li></ul>
  9. 9. Algorithm <ul><li>Clustering: k-Means, get set of k clusters </li></ul><ul><li>“ Document” C is bag of all tags in cluster </li></ul><ul><li>For each tag in C calculate: </li></ul><ul><ul><li>TF = |P(C,t)| </li></ul></ul><ul><ul><li>IDF = |P(R)| / |P(R, t)| </li></ul></ul><ul><ul><li>UF = |U(C,t)|/|U(C)| </li></ul></ul>
  10. 10. Scoring <ul><li>Score (t) = TF * IDF * UF </li></ul><ul><li>Threshold values </li></ul><ul><ul><li>30+ photographs </li></ul></ul><ul><ul><li>Minimum 3 users </li></ul></ul><ul><ul><li>Score > 1 </li></ul></ul><ul><li>Final dataset: (tag, score, latitude, longitude) </li></ul>
  11. 11. DEMO
  12. 12. Precomputation <ul><li>Divide the world into equal sized non-overlapping tiles </li></ul><ul><li>Compute and store the tags for each tile </li></ul><ul><li>Repeat for different zoom </li></ul><ul><li>levels </li></ul>
  13. 13. Retrieval <ul><li>Find the tile level closest in size to the request area </li></ul><ul><li>Select the tiles that fully cover the request area </li></ul><ul><li>Return the tags that fall within the request area </li></ul>
  14. 14. User Study <ul><li>10 subjects </li></ul><ul><li>6 female, 4 male </li></ul><ul><li>Ages 20-60 </li></ul><ul><li>Varying technical knowledge </li></ul><ul><li>No geotagged photos of their own </li></ul>
  15. 15. Experiment tasks <ul><li>Vacation recap </li></ul><ul><li>San Francisco tour </li></ul><ul><li>Explore a new city </li></ul>
  16. 16. Recall <ul><li>Reminded the subject about locations </li></ul><ul><li>“It brings out memories” </li></ul><ul><li>“Oh my God! This place has the best restaurants” </li></ul><ul><li>“We wanted to see the Polynesian Cultural Center&quot; </li></ul>
  17. 17. Discovery <ul><li>Participants discovered previously unknown locations and events </li></ul><ul><ul><li>“I’ve never heard of this festival” </li></ul></ul><ul><ul><li>“There is car racing which I'd probably go see” </li></ul></ul>
  18. 18. Needle & Haystack <ul><li>Excellent visualization of the Haystack </li></ul><ul><li>Hard to find specific information </li></ul><ul><ul><li>“Where was Culver City again?” </li></ul></ul><ul><li>No way to search </li></ul><ul><ul><li>“I guess what I’m looking for are bull fighting pictures” </li></ul></ul>
  19. 19. Other Responses <ul><li>Gets the “vibe” of a place </li></ul><ul><li>Share with other people </li></ul><ul><li>Tags did not always match the mental model of a location </li></ul><ul><li>Wanted more tags </li></ul><ul><li>Want more info about tags </li></ul>
  20. 20. Conclusions <ul><li>Extracted meaningful aggregate information from georeferenced data </li></ul><ul><li>Allows users to explore locations in a new way </li></ul><ul><li>Users like using the overview but also want the ability to search </li></ul>
  21. 21. Future work <ul><li>Adding search capability </li></ul><ul><li>Show photos in places with no tags </li></ul><ul><li>Differentiate locations and events </li></ul><ul><li>Apply to other types of georeferenced data </li></ul>
  22. 22. tagmaps.research.yahoo.com <ul><li>World Explorer </li></ul><ul><li>Data API </li></ul><ul><li>Visualization toolkit </li></ul><ul><li>Trip Explorer </li></ul><ul><li>Night Explorer </li></ul>
  23. 23. Questions? Rahul Nair [email_address] http://tagmaps.research.yahoo.com

×