Improving Keyphrase Extraction Using LL-Ranking

Svetlana Popova, Vera Danilova, Mikhail Alexandrov, John Cardiff

Abstract: Keyphrases provide a concise representation of the main content of a document and can be effectively used within information retrieval systems. In the paper, we deal with the keyphrase extraction problem when a given number of keyphrases for a text should be extracted. The research is focused on the keyphrase candidates ranking stage. In the domain, the question remains open of whether the keyphrase extraction quality can be improved by putting limits on the number of phrases of different lengths extracted during candidate ranking. We assume that the quality of resulting keyphrases can be enhanced if we introduce \(\underline{L}\)imitations on the number of phrases of specific \(\underline{L}\)engths in the resulting set (LL-ranking strategy). The experiments are performed on the well-known INSPEC dataset of scientific abstracts. The obtained results show that the proposed limitations help to significantly increase the quality of extracted keyphrases in terms of Precision and F1.

External IDs:dblp:conf/csit/PopovaDAC19