Skip to content

Latest commit

 

History

History
86 lines (63 loc) · 2.33 KB

README.md

File metadata and controls

86 lines (63 loc) · 2.33 KB
Raw

Entity disambiguation for Europeana Newspapers

A simple Python library and webservice, that allows named entity disambiguation against a label database. The idea is to use a Solr query to filter possible candidates and use the more detailed analysis on string similarity, number of inlinks and entity type to select the "best" candidate. It contains code to handle (multi-lingual) DBpedia dumps and load them into a Solr backend. It also contains helper code for the annotation of ALTO 2.1 files that are used in the context of the Europeana Newspapers project.

Startup Solr

This project needs a Solr 4 instance. A script to install a local instance is here:

 init\install-solr.sh

If you use this way, you can use the

./start-solr.sh

to start up a local instance with the right configuration

Processing dumps

Download dumps from DBPedia (nt-format) from http://wiki.dbpedia.org/Downloads2014 and put them in data/rawdata

For an English label database you need:

instance_types_en.nt.bz2	
page_links_en.nt.bz2		
short_abstracts_en.nt.bz2
labels_en.nt.bz2		
redirects_en.nt.bz2

Run

cd data
./prepare-solr-input.sh

This will take a while and generate a final.csv, which is the input

To index the CSV file into the default Solr, use

./index-in-solr.sh

Webservice

There is a webservice, that currently only support DBpedia link resolution for a named entity It has a built-in webserver. To start

python disambiugation/web.py

It will listen on port 5000

To test it:

http://localhost:5000/link?ne=einstein

Results in:

  {
	"p": 0.4927993981146857,
	"link": "http://dbpedia.org/resource/Albert_Einstein",
	"ne": "einstein",
	"name": "Albert Einstein"
 }

It uses the Bottle framework, so it should be possible to use the class also in a WSGI environment (Apache). See documentation from the Bottle project.

ALTO processing

There is a script that processes a directory of ALTO files, checks the annotations and adds an 'URI' argument to that annotation if it finds a DBpedia link for that named entity and writes the XML file to an output directory.

For example

python disambiguation/process-alto.py SOURCE-DIRECTORY-TREE OUTPUT-DIRECTORY-TREE en