Adaptive learning for improving semantic tagging of scientific articles

In this paper we consider a problem of automatic labeling of textual data with concepts explicitly defined in an external knowledge base. We describe our tagging system and we also present a framework for adaptive learning of associations between terms or phrases from the texts and the concepts. Those associations are then utilized by our semantic interpreter, which is based on the Explicit Semantic Analysis (ESA) method, in order to label scientific articles indexed by our SONCA platform. Apart from the description of the learning algorithm, we show a few practical application examples of our system, in which it was used for tagging scientific articles with headings from the MeSH ontology, categories from ACM Computing Classification System and from OECD Fields of Science and Technology Classification.