Code supporting my Master Thesis about finding similar neighborhood across cities using social media activity.

You can get more information by reading our blogpost, a two pages academic summary, our ICWSM paper or if you have more time on your hands, my complete thesis.

If you're interested in our dataset, you can find it on Figshare. As Foursquare prohibits distributing out of data venue information, you will have to use FillDB.py to collect the latest statistics about them.

Below I provide more technical details. Yet it should be noted that for now, not all data are included with the code (although the most important can be found on dropbox and thus there is no simple demonstration one can quickly test. Hopefully this will soon be remedied :)

All the code is written in Python 2 (but is known to work with automatic 2to3 modifications under Python 3.4 as well) and dependencies can be install from requirements.txt. Map rendering is controlled by a Flask app and involve Leaflet as well as some javascript in the static directory.

First we collect data from two social media, Foursquare and Flickr. Then we aggregate data at the venue level. Each venue become a feature vector and is stored in a matrix.

Finally we devise a method that given a polygon in one city, compute the $k$ most similar in another city.

• grab_photos.py retrieve a list of all Flickr photos taken in a given city and insert them with additional metadata in a mongo database

• twitter.py Listen to public twitter stream for Foursquare checkin.
• boost_twitter.py Gather more tweets by fetching timeline of previously discovered users

• AskFourquare.py Parse JSON Foursquare responses to relevant python object
• CheckinAPICrawler.py Get checkin info by requesting Foursquare API. Unfortunately, it's not working since Foursquare deployed its Swarm application, even though it can be modified to handle this new case
• FillDB.py Use collected tweet and AskFourquare to request information about users and venues, before inserting them in a Mongo database
• FSCategories.py Maintain a tree of Foursquare categories and provide query methods

• VenueFeature.py This main step is to transform the raw data collected into a feature matrix whose rows are each venue of a city with enough visits and column are the features described in table 3 page 27.
• Surrounding.py Maintain a KD tree of venues to allow spatial ball query
• FlickrVsFoursquare.py Compute discrepancy between tweets and photos. It's not a feature associated with a single venue so it was not used in the thesis but it's interesting nonetheless. Basically, it divides a city into a grid and find the cell where the proportion of check-ins and photos are unusual. This discriminates between very touristic location that are mostly photographed (Eiffel Tower) against location where things happen (Stadium, Railway station, …)

• worldwide.py Defines query_in_one_city, which performs a single similarity query between a GeoJSON polygon from one city to another
• one_approx_query.py Illustrate the use of worldwide.py on a predefined set of queries
• approx_emd.py Avoid computing the Earth Mover's Distance between all possible rectangles in the target city by using the pruning strategy described in section 6.1 page 38
• ClosestNeighbor.py Perform $k$-nearest neighbor queries over venues in two cities
• neighborhood.py This one does too many things and most of them turned to be not working anyway

• ServeNN.py A flask webserver with the following interesting routes
  • /n/<origin>/<dest> offer a selection of neighborhood
  • /<origin>/<dest>/<int:knn> interactively pick a venue in the origin city and show its $k$ nearest neighbors in the dest city

• cities.py Define the 20 cities we choose by their bounding box and provide methods to convert between latitude, longitude and local euclidean coordinates.
• arguments.py
• calc_tsne.py
• Chunker.py
• clean_timeline.py
• CommonMongo.py
• Counter.py
• LocalCartesian.py
• OrderedDict.py
• persistent.py
• RequestsMonitor.py
• explore.py
• twitter_helper.py
• utils.py

• 0708tue.py
• 0813wed_map.py
• alt_emd.py
• bench.py
• CheckinCrawler.py
• cluster_city.py
• common_tag.py
• compare_tags.py
• compile_cython.py
• CorrectCheckIn.py
• extract_dataset.py
• extract_gold.py
• figure4.py
• first_query.py
• gen_status.py
• geom_stat.py
• get_brands.py
• ir_evaluation.py
• LDA.py
• learn_weights.py
• local_vs_tourist.py
• merge_gold.py
• more_query.py
• nldm.py
• outplot.py
• places_and_venues.py
• plot_corr.py
• plot_tag.py
• preprocess.py
• ProgressBar.py
• rank_disc.py
• read_foursquare.py
• report_metrics_results.py
• saved3.py
• seetags.py
• selection.py
• significance_test.py
• spatial_scan.py
• specific_emd_dst.py
• tag_support.py
• time_all_cities.py
• top_metrics_circle.py
• VenueIdCrawler.py
• wordplot.py
• emd_leftover.py