Retrieving Complex Objects with HySpirit

Traditional Information Retrieval (IR) considers documents as atomic units. In this paper, we show the retrieval of the components of the documents which satisfy best the information need. This finer granularity eases the browsing of the retrieval result. The approach supports multimedia and networked IR since multimedia documents are composed of other objects and networks combine several collections comprising the documents. We gain a unified view on networks, databases, and multimedia documents by considering them as complex objects — retrieval among a heterogeneous document corpus can be modeled appropriately. We present a probabilistic retrieval function where the initial estimation of probabilistic parameters is based on the logical structure of documents and the retrieval process is described as probabilistic logical inference. Probabilistic parameters and the retrieval process are represented in probabilistic Datalog programs which are executed by HySpirit — a system for processing probabilistic inference.