News

We are at ESWC 2018 - Entity Linking in 40 Languages using MAG

Project page: http://aksw.org/Projects/AGDISTIS.html
Source code: https://github.com/AKSW/AGDISTIS
Manual: https://github.com/AKSW/AGDISTIS/wiki

 

A recent survey by IBM  suggests that more than 2.5 quintillion bytes of data are produced on the Web every day. Despite the complexity of the task, Entity Linking (EL) approaches have recently achieved increasingly better results by relying on trained machine learning models. A portion of these approaches claim to be multilingual and most of them rely on models which are trained on English corpora with cross-lingual dictionaries. However, MAG (Multilingual AGDISTIS) shows that the underlying models being trained on English corpora make them prone to failure when migrated to a different language. Additionally, these approaches hardly make their models or data available on more than three languages. To this end, we released a new version of MAG which supports 40 different languages using sophisticated indices. Additionally, we provide a Wikidata index in order to prove MAG’s agnosticism.

Our work was accepted at the Demo and Posters session at ESWC 2018. It is running at http//agdistis.aksw.org/mag-demo.
 

Where can I find the indices?

All indices can be found at https://hobbitdata.informatik.uni-leipzig.de/agdistis/. When making it run, you must pay attention to the properties file as setting the URIs (baseURI, nodeType and edgeType) correctly helps MAG perform properly.
 

STAY TUNED!

____________________________________

  1. tinyurl.com/ibm2017stats
  2. tinyurl.com/y9m9w89a
  3. tinyurl.com/agdistis-properties
527efb333