hyProximity

Abstract

As opposed to existing keyword suggestion systems (such as AdWords Keyword Tool) that suggest keywords based on co-occurrence in searchers, in texts etc. we propose a system that uses DBPedia to help the user discover (potentially unknown) relevant keywords. Our system thus suggests keywords based on their meaning, and conceptual relations, rather then based on co-occurring use of the keywords. Our approach consists of (1) finding the DBPedia concepts that are directly referred to in the user's text, using Zemanta API; and then (2) using the graph of skos:broader relations in DBPedia to find new concepts that are on a short distance from the concepts originally found in text (called seed concepts). Our technique gives higher priority to the concepts that are found in a short distance surrounding of several seed concepts then those that are close to only one seed concepts. This feature allows us to identify the context of relevancy (for instance a concept can have a number of neighboring concepts in skos;broader graph, but only some of them are relevant for the context of the given text - the others are relevant but for some other context: e.g., clay is close to mineral extraction and to use of clay in medical industry; but the later is irrelevant for the text dealing with mineral extraction machinery ).