Connecting Words and Linked Data Concepts by Latent Features

Authors: Márius Šajgalı́k, Michal Barla, Mária Bieliková, Julian Szymański
Year: 2015
Venue: 1st International KEYSTONE Conference
Link: http://www.keystone-cost.eu/ikc2015/
Product of the Action: Yes

Keystone Members Authors:
, ,

Abstract:
We are now flooded with vast number of linked data sources focused on wide range of domains. On the other hand, unsupervised learning of latent word features has come into attention, which shows that it is possible to learn meaning of words from raw unlabelled and unstructured text. In this paper, we present our vision of combining these two approaches into a single one, which would simplify management of linked data, enable us to automatically learn meaning of words, concepts and relations between them, identify occurrences of concepts in the text and learn new emerging concepts just from their descriptions.