Stand-alone ranking systems based on highly optimized inverted file structures are generally considered `the' solution for building search engines. Observing various developments in software and hardware, we argue however that IR research faces a complex engineering problem in the quest for more flexible yet efficient retrieval systems. We propose to base the development of retrieval systems on `the database approach': mapping high-level declarative specifications of the retrieval process into efficient query plans. We present the Mirror DBMS as a prototype implementation of a retrieval system based on this approach.
Arjen P. de Vries