Challenges for Industrial-Strength Information Retrieval on Databases

Implementing keyword search and other IR tasks on top of relational engines has become viable in practice, especially thanks to high-performance column-store technology. Supporting complex combinations of structured and unstructured search in real-world heterogeneous data spaces however requires more than “just” IR-on-DB. In this work, we walk the reader through our industrial-strength solution to this challenge and its application to a real-world scenario. By treating structured and unstructured search as first-class citizens of the same computational platform, much of the integration effort is pushed from the application level down to the data-management level. Combined with a visual design environment, this allows to model complex search engines without a need for programming.