Towards a practical semantic search - case study on the TV guide domain

Ad-hoc semantic search aims at the retrieval from semantic data using queries formulated in a natural language. In this paper we explore several approaches to the ad-hoc semantic search and study their behaviour and quality of results in the scope of a domain specific application. The application concerns with the search in the program listings of TV stations. The goal is to find relevant program items, given a query formulated in natural language, or by using keywords. The queried data is a semantic knowledge base containing the data on the program listings as well as detailed information on distinct program items. We explore two approaches to the entity modelling and entity search, the first is based on traditional information retrieval methods and is inspired by the recent developments in the field. The second methods exploit deeper query analysis, query mapping to entities of the underlying knowledge base and the graph topology exploration to identify answers to the input query. The evaluation of the two approaches shows that the second method outperforms the existing approach, having the mean average precision almost two times higher than the former one. The results indicate that for a well specified, closed domain, ad-hoc semantic retrieval models should go beyond the traditional information retrieval approaches.