Keyword detection in natural languages and DNA

Abstract

We show that words in a text present long-range frequency fluctuations due to a strong self-attraction, that is directly related to the relevance of the term to the text considered. The standard deviation of the distance between successive occurrences of a word is an excellent parameter to quantify this self-attraction, and provides us with an effective tool for automatic keyword extraction. DNA sequences also present the same features: "words", for example codons in the coding part of the sequences, attract between themselves.