DIKEA: Domain-Independent Keyphrase Extraction Algorithm

David X. Wang, Xiaoying Gao, and Peter Andreae

Abstract. This paper introduces a new domain-independent keyphrase extraction system (DIKEA). Keyphrase extraction is a challenging problem that automatically extracts or assigns keyphrases to documents and it can benefit many research areas such as information retrieval, particularly indexing, clustering, and summarization. A landmark research KEA (Keyphrase Extraction Algorithm) formulated the problem as a supervised machine learning problem and successfully applied a Naïve Bayes model to it, which showed great promise but the performance is not satisfactory. Its state-of-the-art extension KEA++ has a significantly improved performance but relies on a domain specific vocabulary which is often not available or not complete. This paper introduces a novel domain-independent approach and has three main contributions: utilising the largest online knowledge source—Wikipedia—for keyphrase candidate selection; presenting new features for keyphrase evaluation, including a Wikipedia-based feature–link probability; and evaluating a number of different learning algorithms, including multilayer perceptrons, for keyphrase selection. Experiments show that our system clearly outperforms KEA and closely matches the performance of KEA++, without requiring any domain-specific knowledge such as KEA++’s vocabulary list.

LNAI 7691, p. 719 ff.

Full article in PDF | BibTeX

lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2012