Indri : A language-model based search engine for complex queries ( extended version )

Search engines are a critical tool for intelligence analysis. A number of innovations for search have been introduced since research with an emphasis on analyst needs began in the TIPSTER project. For example, the Inquery search engine introduced support for specification of complex queries in a probabilistic inference network framework. Recent research on language model-ing has led to the development of Indri, a search engine that combines the best features of inference nets and language modeling in an architecture designed for large-scale applications. In this paper, we describe the Indri system and show how the query language is designed to support modern language technologies. We also present results demonstrating that Indri is both effective and efficient.