This project contains files exploring a co-occurrence frequency analysis of Tumblr tags as an investigation of the ways that community drives language usage by using co-occurence as my metric of relatedness to prevent introducing researcher bias through the application of a hypothetical 'golden standard' of relatedness.
In plain English, what that means is that I'll be looking into the ways in which users of Tumblr use tags together as a way of understanding those words relatedness. By looking at the ways users themselves define and use words together I hope to avoid forcing language usage into my own understanding and to allow multiple forms of 'correct' word usage to exist simultaneously in the same system.
This project has been built using Java, Javascript, Python2.7, and a tiny bit of C
You can expect to see a heavy usage of the SciKitLearn, PyTumblr, and D3.js libraries
SciKitLearn and PyTumblr can both be pip installed
Basically I wanted to see if I could do unsupervised machine learning feature extraction by deconstructing and reimagining the way that K-Means works as a way to asymptotically approach a specific subset of tags that all shared similar meanings
Since language is so community driven in the TumblrVerse I was hoping that these breaks would fall along community lines and that further visualizations would visualize the ways in which communities interact.
##TODO:
- Push up Java files so it will run again
- Look into python foreign language injection and using pthreads in C to speed everything up
- Figure out how to separate overlapping tags in javascript for greater legibility in the visualizer
- Keep track of which users a tag belongs to as a way to further visualize clustering success
- Color code tag tesselations based on which user said that tag most
- Interface with Tumblr API to generate a map for a single user to show how they enteract with the network