Research on weighted complex network based keywords extraction

Abstract

Based on the complex network theory, this paper constructs a weighted lexical network to extract keywords from the text automatically. The current related researches mainly focus on the measures of nodes' contribution to the whole network, while this paper lays emphasis on the construction of lexical network. By introducing linguistic knowledge, we center on reasonable selection of nodes, proper description of relationships between words, enhancement of node attributes, and etc. Experiments indicate that the lexical network constructed by our approach achieves preferable effect on accuracy, recall and F-value - when selecting the top three results, the three indices increase by 6.67%, 3.96% and 4.97% on average than the classic TF-IDF method respectively. © 2013 Springer-Verlag.