Random Walks on Text Structures

Graph Ranking on Maximal Frequent Sequences for Single Extractive Text Summarization

“…Particularly relevant in this paper is the application of random walks to text processing, as done in TextRank system [7]. TextRank has been successfully applied to three natural language processing tasks [8]: document summarization [3; 7], word sense disambiguation [9], and keyword extraction, and text classification [10] with results competitive with those of state-of-the-art methods.…”

Terms Derived from Frequent Sequences for Extractive Text Summarization

“…Song et al [3] basically weight a word basing on the number of lexical connections, such as semantic associations expressed in a thesaurus, that the word has with its neighboring words; along with this, more frequent words are weighted higher. Mihalcea [15] presents a similar idea in the form of a neat, clear graph-based formalism: the words that have closer relationships with a greater number of "important" words become more important themselves, the importance being determined in a recursive way similar to the PageRank algorithm used by Google to weight webpages.…”

“…The latter idea can be applied directly to sentence weighting without term weighting: a sentence is important if it is related to many important sentences, where relatedness can be understood as, say, overlap of the lexical contents of the sentences [15]. The two methods presented in [15] are those that currently give the best results and with which we compare our suggested method.…”

“…The two methods presented in [15] are those that currently give the best results and with which we compare our suggested method.…”

“…-State of the art: The author of [15] provided us with her data, which were evaluated in the same conditions as proposed methods. Specifically, DirectedBackward version of TextRank [15] was evaluated.…”