TV navigation agent for measuring semantic similarity between programs

Abstract

This paper proposes a method of computing semantic similarity which improves accuracy compared to the exisitng approaches. Most of the exisitng approaches uses WordNet as the ontology to calcualte and evaluate the similarly. A method using lightweight ontologies is proposed. The proposed method which considers offset value depends on ontology's hierarchy layer and keyword's importance increases the accuracy of the correlation between computed similarities and human judgements.The method was used in a TV navigation agent system. This system introduces similar TV programs in which the user is interested. Several lightweight ontologies for TV programs to compute semantic similarity between TV programs was also developed. An experiment to evaluate the accuacy was also conducted. © Springer-Verlag Berlin Heidelberg 2007.