Case-based reasoning meets information retrieval
Malika Smaïl, Marion Créhange · Multimedia Information Retrieval · 1994
No single mechanism can be expected to give optimal retrieval in all cases in information retrieval (IR) context. Many problems have been reported concerning the difficulty of making the right choices at the design stage of information retrieval systems. This paper presents a flexible model of information retrieval process together with a mechanism exploiting query typologies for instantiating the flexible model. The proposed model integrates in a single framework various alternative information retrieval primitives. This is possible thanks to attaching to every primitive strategic parameters which express different alternatives. We show how the parameterized model can be combined with a Case-Based Reasoning (CBR) approach to incrementally improve an information retrieval strategy. The proposed approach fulfils a synergy between CBR and information retrieval which aims to exploit users feedback for improving the retrieval short-term (during a single retrieval session) and long-term performances (over the life time of the system). This approach allows incremental learning of the right use of a generic model as well as exploiting success or failure of information retrieval. This is achieved by managing a memory of prior search sessions. A working prototype exists and evaluation is in the initial phase. However we anticipate some difficulties during the evaluation due to interactivity in IR process and we give a prospective study of how the performances will improve.