World Explorer (JCDL 2007 Best Paper)

1yahooresearchberkeley.comRahul Nair - World Explorer

World Explorer: Visualizing Aggregate Data from Unstructured Text in Geo-

Referenced Collections

Shane Ahern, Mor Naaman, Rahul Nair* & Jeannie Yang

Yahoo! Research Berkeley


Attraction Map of Paris

Stanley Milgram, 1976. Psychological Maps of Paris


Attraction Map of Paris

Y!RB, 2007.


Flickr “geotagged”

20+ million images

Can we do better?


Location-driven Modeling

• Derive meaningful data about map regions

• E.g., representative tags, photos


Data Description


Issues

• Sparse data set• Photographer bias

– In location– In tags

• Incorrect data


Heuristics

• Number of photographs denotes the “importance” of a location

• Users will use a common subset of tags to describe objects/locations

• Concentrated tag usage indicates descriptiveness


Algorithm

• Clustering: k-Means, get set of k clusters

• “Document” C is bag of all tags in cluster

• For each tag in C calculate: – TF = |P(C,t)|– IDF = |P(R)| / |P(R, t)|– UF = |U(C,t)|/|U(C)|


Scoring

• Score (t) = TF * IDF * UF• Threshold values

– 30+ photographs – Minimum 3 users– Score > 1

• Final dataset: (tag, score, latitude, longitude)

11

DEMO


Precomputation

• Divide the world into equal sized non-overlapping tiles

• Compute and store the tags for each tile

• Repeat for different zoom levels


Retrieval

• Find the tile level closest in size to the request area

• Select the tiles that fully cover the request area

• Return the tags that fall within the request area


User Study

10 subjects• 6 female, 4 male• Ages 20-60• Varying technical knowledge• No geotagged photos of their own


Experiment tasks

• Vacation recap• San Francisco tour• Explore a new city


Recall

Reminded the subject about locations• “It brings out memories”• “Oh my God! This place has the best

restaurants”• “We wanted to see the Polynesian

Cultural Center"


Discovery

Participants discovered previously unknown locations and events– “I’ve never heard of this festival”– “There is car racing which I'd probably

go see”


Needle & Haystack

• Excellent visualization of the Haystack• Hard to find specific information

– “Where was Culver City again?”

• No way to search– “I guess what I’m looking for are bull

fighting pictures”


Other Responses

• Gets the “vibe” of a place• Share with other people• Tags did not always match the mental

model of a location• Wanted more tags• Want more info about tags


Conclusions

• Extracted meaningful aggregate information from georeferenced data

• Allows users to explore locations in a new way

• Users like using the overview but also want the ability to search


Future work

• Adding search capability• Show photos in places with no tags• Differentiate locations and events• Apply to other types of georeferenced

data


tagmaps.research.yahoo.com

• World Explorer• Data API• Visualization toolkit

• Trip Explorer• Night Explorer

23

Questions?

Rahul [email protected]

http://tagmaps.research.yahoo.com

World Explorer (JCDL 2007 Best Paper)

Technology

photos yahooresearchberkeley

representative tags

new city yahooresearchberkeley

aggregate data

common subset of tags

map regionse

comattraction map of

tagsdifferentiate locations