ccs

Content Classification System - infer semantic information about unstructured data

Content Classification System

This is an experiment in classifying content based on unstructured properties of that content. It requires a set of "seed" data to provide some statistical information about the distribution of data points over the previously classified "seed" data.

Right now the system operates by analysing the frequency of words in the title of torrents and the categories assigned to them. It constructs a list of words and the number of times they have been seen in each category. To guess the category of an unknown title, it adds up the (normalised) frequencies of use of each word in the title for each category and presents the result as a list of probably categories in descending order.

git clone git://github.com/deoxxa/ccs.git
cd ccs
npm install
MONGODB=127.0.0.1:27017/ccs ./grabber.js
MONGODB=127.0.0.1:27017/css ./mapreduce.js
MONGODB=127.0.0.1:27017/css ./search.js "[Silent-Raws] Queen's Blade ~Rebellion~ - 01 (AT-X 1280x720 x264 AAC).mp4"

BSD, 3-clause. A copy is included.