Associative Search Techniques versus Probabilistic Retrieval Models

This article offers a personal look back at the origins and early use of associative search techniques, and also a look forward at more theoretical approaches to the document retrieval problems. The purpose is to contrast the following two different ways of improving system performance: (1) appending associative search techniques to more or less standard (conventional) document retrieval systems, and (2) designing document retrieval systems based on more fundamental and appropriate principles namely probabilistic design principles. Very recent work on probabilistic approaches to the document retrieval problem has provided a new (and rare) unification of two previously competing models. In light of this, I argue that if we had to choose the best way to improve performance of a document retrieval system, it would be wiser to implement, test, and evaluate this new unified model, rather than to continue to use associative techniques which are coupled to conventionally designed retrieval systems.