Combining word embeddings as a tool for subject identification

Abstract

This talk shows ungoing work aiming at finding subject matter relations between text documents and word clouds. A number of increasingly successful semantic word embedding procedures - learning semantic relations from contextual distributions - have been developed in recent years. We are interested in using these embeddings for mapping documents to subjects without having to resort to supervised training of classifiers. In order to improve the quality of such a mapping it seems necessary to combine several embedding strategies. Some approaches in this direction are presented.