A new TV program retrieval method using a semantic relations dictionary

Broadcasting companies have started offering video-on-demand services to provide their viewers with various kinds of TV programs. Most of these services have a program retrieval system that helps users find the TV programs they would like to watch. The conventional systems retrieve TV programs by keyword matching of the query and the TV program summary. However, it is difficult for these systems to retrieve TV programs from small databases (e.g. databases where the broadcasting companies select the programs) because there are only a few programs featuring summaries that include the query. As a result, retrieval systems often output very few programs. To tackle this problem, we propose a new retrieval method using a semantic relations dictionary (SRD). We first generate a graph structure from the SRD and TV programs and then calculate the similarity scores between an input query and all TV programs. We use four metrics - the distance in the graph structure, the number of paths from query to TV program, word similarity and word ambiguity - to calculate similarity scores effectively and efficiently. The system then retrieves TV programs that have a high similarity score with the query.