Semanticrank: Ranking keywords and sentences using semantic graphs

The selection of the most descriptive terms or passages from text is crucial for several tasks, such as feature extraction and summarization. In the majority of the cases, research works propose the ranking of all candidate keywords or sentences and then select the top-ranked items as features, or as a text summary respectively. Ranking is usually performed using statistical information from text (i.e., frequency of occurrence, inverse document frequency, co-occurrence information). In this paper we present Semant- icRank, a graph-based ranking algorithm for keyword and sentence extraction from text. The algorithm constructs a semantic graph using implicit links, which are based on semantic relatedness between text nodes and consequently ranks nodes using different ranking algorithms. Comparative evaluation against related state of the art methods for keyword and sentence extraction shows that Semanti- cRank performs favorably in previously used data sets.