A new network model for extracting text keywords

Text keywords are defined as meaningful and important words in a document, which provide a precise overview of its content and reflect the author's writing intention. Keyword extraction methods have received a lot of attentions, among which is the network-based method. However, existing network-based keyword extraction methods only consider the connections between words in a document, while ignoring the impact of sentences. Since a sentence is made of many words, while words affect one another in a sentence, neglecting the influence of sentences will result in the loss of information. In this paper, we introduce a word network whose nodes represent words in a document, and define that any keyword extraction method based on a word network is called as a Word-net method. Then, we propose a new network model which considers the influence of sentences, and a new word-sentence method based on the new model. Experimental results demonstrate that our method outperforms the Word-net method, the classical term frequency-inverse document frequency (TF-IDF) method, most frequent method and TextRank method. The precision, recall, and F-measure of our result are respectively 7.95, 8.27 and 6.54% higher than the Word-net result, and the average precision of our result is 17.56% higher than the TF-IDF result. A two-way analysis of variance is employed to validate the empirical analysis, which indicates that keyword extraction methods and keyword numbers have statistically significant effects on the evaluation of metric values.