Automated subject indexing using word embeddings and controlled vocabularies: a comparative study

Abstract

Text mining methods contribute significantly to the understanding and the management of digital content, increasing the potential of entry links. This paper introduces a method for subject analysis combining topic modelling and automated labelling of the generated topics exploiting terms from existing knowledge organisation systems. A testbed was developed in which the Latent Dirichlet Allocation (LDA) algorithm was deployed for modelling the topics of a corpus of papers related to the Digital Library Evaluation domain. The generated topics were represented in the form of bags-of-words word embeddings and were utilised for retrieving terms from the EuroVoc Thesaurus and the Computer Science Ontology (CSO). The results of this study show that the domain of DL can be described with different vocabularies, but during the process of automatic labelling the context needs to be taken into account.