RDFSpace

This Python library helps generating a low-dimensional vector space from very large RDF graphs in a reasonable time. For example it is possible to process large sections of DBpedia on a commodity laptop. Once this space is generated, it can be used to compute fast similarities between URIs, or to compute the Eigenvector Centrality (~pagerank) of URIs.

Getting started

Setting up:

$ apt-get install python-pip python-librdf python-numpy python-scipy python-nose

Installing:

$ python setup.py install
# (You might have to run that twice to get around https://github.com/piskvorky/sparsesvd/pull/4)

Running the tests:

$ nosetests

Or from pypi:

$ pip install rdfspace

Example use

$ cd examples
$ gunzip influencedby.nt.gz
$ python
>>> import rdfspace
>>> from rdfspace.space import Space
>>> space = Space('influencedby.nt', ignored_predicates=['http://www.w3.org/1999/02/22-rdf-syntax-ns#type'], rank=50)
>>> space.similarity('http://dbpedia.org/resource/JavaScript', 'http://dbpedia.org/resource/ECMAScript')
>>> space.centrality('http://dbpedia.org/resource/JavaScript')
>>> space.similarity('http://dbpedia.org/resource/Albert_Camus', 'http://dbpedia.org/resource/JavaScript')
>>> space.centrality('http://dbpedia.org/resource/Albert_Camus')
>>> space.similar('http://dbpedia.org/resource/Albert_Camus')

How it works

RDFSpace construct a sparse adjacency matrix from an input RDF file. We perform Singular Value Decomposition on this sparse adjacency matrix to approximate this space, which gives us a lower-dimensional space capturing URI similarities. This space can then be used for a wide range of uses, e.g. automated tagging, disambiguation, etc.

Licensing terms and authorship

See 'COPYING' and 'AUTHORS' files.

All code here, except where otherwise indicated, is licensed under the GNU Affero General Public License version 3. This license includes many restrictions. If this causes a problem, please contact us. See "AUTHORS" for contact details.