Multi-class named entities extraction from biomedical literature

Abstract

With rapid growth of electronic literature in recent years, efficient named entities extraction becomes an indispensable part of knowledge base construction automation. In this paper an entity extraction system useful as biomedical knowledge acquisition was presented. Unlike most entity extraction systems which do not concern term variants, the proposed system was incorporated with a rule-based resolver to recover the full forms of those target entities from the coordination variants. The resolution approach was proved with GENIA Corpus 3.0 to be feasible by showing 88.51% recall and 57.04% precision. On the other hand, the kernel part of the system was based on Hidden Markov Model (HMMs) by setting appropriate set of input features extracted from training corpus. With various experiments on different corpora the proposed system achieved promising results at entity boundary identification and at classification as well.