Keywords Extraction Based on Word Relevance Degrees

Abstract

Keywords extraction (KE) is an important part of many neural language processing (NLP) tasks which have attracted much attention in recent years. Graph-based KE methods have been widely studied because it is always unsupervised and can extract keywords with information among words. However, existing graph-based KE methods suffer from low time efficiency or large corpus dependency. In this work, we propose a new graph-based keywords extraction method which uses word relevance degrees to extract keywords and two word relevance degrees calculation algorithms. The proposed method doesn't rely on big corpus and experimental results show that the proposed method can extract keywords more efficient with higher performance on compared with TF-IDF, TextRank and KMST methods.