Skip to content
Code for Joseph, Carley and Hong (2014)
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
mutual3 Added first version of the code Aug 9, 2013

This is a trimmed, more readable version of the analysis code used in the article:

  title={Check-ins in “Blau space”: Applying Blau’s macrosociological theory to foursquare check-ins from New York City},
  author={Joseph, K. and Carley, Kathleen M. and Hong, Jason I.}
   journal={ACM Transactions on Intelligent Systems and Technology (TIST)},

Please refer to this article if you use the code.

However, please note that code was adopted from the following sources:

  1. MALLET tutorial:

  2. Normalized Mutual Information code:

  3. Haversine distance:

  4. Walker random sampling class:

If you use [1] or [2], please also cite the related articles (found at the websites listed)

This is in addition to all of the amazing libraries used in Java, Python and R and the use of the census API to pull down the demographic data.

The code is unfortunately not explicitly runnable because I can't give away the data and, possibly, because I used way too many absolute paths and I might have made a mistake cleaning them up . Apologies in advance, but I've done some work to make the code more readable. If you actually want to use pieces, shoot me an email and I can clarify.

The process used by paper was as follows:

  1. Pull the data out of the mysql database where it was stored. Check-in data is then in the form
2011-08-07 00:27:45  31862406	4b9add62f964a52078dd35e3	Food::Mexican Restaurant	-74.032	40.6218	Trace
  1. Run "process.R", which did the following: -generated cleaned check-in files and a file that included demographic information for all of the venues -ran the experiments with the different parameters of LDA -ran the NMI code to find the agreement between the two data partitions -generated the figures in the article

  2. There is a comment in process.R at the point at which was run to generate the bootstrapped CIs

The LDAs are run using a little Java program that interfaces with the MALLET toolkit. I've provided the jar that's used in process.R as well as the source code.

The file lda_experiment.csv holds the results of the experiment run to determine which parameterization of LDA to use

The file topicWordWeights5_5_1_true_20_0.01_0.6_5.txt holds the output of the LDA used in the analysis in the article.

You can’t perform that action at this time.