Modelling RDF data as a vector space

## Project description

This Python library helps generating a vector space from very large hierarchies encoded in RDF. An obvious example application is to generate a vector space from a SKOS hierarchy or an RDFS subclass hierarchy.

## Getting started

Running the tests:

$nosetests Installing:$ python setup.py install

## Example use

$wget http://downloads.dbpedia.org/3.7/en/skos_categories_en.nt.bz2$ bunzip2 skos_categories_en.nt.bz2 \$ python >>> from rdfsim.space import Space >>> space = Space(‘skos_categories_en.nt’) >>> space.similarity_uri(category1, category2)

Constructing a vector space for the entire DBpedia SKOS category hierarchy (3M triples) takes a couple of minutes on a commodity laptop, and has a memory footprint of about 500M.

Alternatively, a subset of it is available in the examples/ directory.

## How it works

For each topic t in the hierarchy, we consider the set of its parents parents(t, k) at a level k. We construct a vector for each t in a space where each dimension corresponds to a topic d in the hierarchy. The value of t on dimension d is defined as follows:

t_d = sum_{k = 0}^{max_depth} sum_{d in parents(t, k)} decay^k

where max_depth and decay are two parameters, which can be used to influence how much importance we attach to ancestors that are high in the category hierarchy.

They can be specified as follows:

>>> Space.max_depth = 8
>>> Space.decay = 0.9


## Licensing terms and authorship

See ‘COPYING’ and ‘AUTHORS’ files.

## Project details

Uploaded source