Improving Semantic Search in Digital Libraries Using Multimedia Analysis

Semantic search of cultural content is of major importance in current digital libraries, such as in Europeana. Content metadata constitute the main features of cultural items that are analysed, mapped and used to interpret users' queries, so that the most appropriate content is selected and presented to the users. Multimedia, especially visual, analysis, has not been a main component in these developments. This paper presents a new semantic search methodology, including a query answering mechanism which meets the semantics of users' queries and enriches the answers by exploiting appropriate visual features, both local and MPEG-7, through an interweaved knowledge and machine learning based approach. An experimental study is presented, using content from the Europeana digital library, and involving both thematic knowledge and extracted visual features from Europeana images, illustrating the improved performance of the proposed semantic search approach.