Introducing
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
Habituses & Habitat:
Behavior-based user classification...
Rapid transit users
Southwest population
Moscow hipsters
Travelers
80%
List of all somehow significant venues represents almost complete list of hipster's "must" places
16%
of significance concentrated on airports and railway stations.
of summary significance concentrated on venues, situated between Kutuzovsky and Leninsky avenues
of significance concentrated on train stations
42%
of significance concentrated on subway stations
...and new urban topology
4sq-MSK habitat map
User's venues choice
Problems & plans
After flattening venues LDA clusterisation we can build habituses-based map of Moscow hyperspace.
Hipsterburg
Travelfield
First experiment shows that LDA model can be used to classify foursquare users, but:
- Venues classification based on ~400 most popular venues is like a text classification based on most common words - it's too general.
- Due to API rate limits every venue was checked once an hour. Supposedly, significant piece of data was lost.
More data about venue's statistics could improve classification and mapping quality.
Another way of classification could be based on time-series data. It also require more statistics.
These classifications could be used to build non-trivial recommender system.
- User's venue choice is always influenced by some causes, sometimes unconscious; e.g. geographic neighbourship, life-style appropriateness, or conspicuous consumption.
- We can represent every user as a text in bag-of-words model where venues are words, and try some topic-modeling techniques.
- E.g., LDA model provides an opportunity for both venues and users classification.
- Different user-choice patterns, based on hidden causes ("topics" in LDA model), form habituses - structuring structures, systems of practice and tendencies.
- Aggregated on the basis of habituses venues form habitat - hyperspace with new, non-euclidean topogy.
- To test all these assumptions I used one-month "herenow" statistics for ~400 trending Moscow venues.
- Here represented most intresting and interpreted habituses.
Subwayland
Nightlifedale
Shoppingham
Southwestington
Alexander Tolmach
http://www.facebook.com/atolmach
zoom&explore: http://zoom.it/MGaI