Sub-symbolic approaches to information retrieval
Paul V. Biron, Donald H. Kraft · 2002
As a result of the knowledge representations and inferencing techniques used in traditional approaches to information retrieval, previous work has experienced extreme difficulty with fundamental aspects of the information retrieval problem: namely, the ambiguity and uncertainty involved in the relevance relationship between users information needs and documents which might satisfy those needs. Previously, we reviewed recent work in the use of connectionist models and genetic algorithms for relevance feedback. In this work, the underlying mechanisms of connectionist and genetic models, parallel sub-symbolic computation, are explored in more detail. Three characteristics of sub-symbolic approaches are singled out as particularly appropriate for addressing the problems posed by ambiguity and uncertainty: (1) robustness; (2) constructivity; and (3) adaptivity.