Towards Dynamic Maintenance of Retrieval Knowledge in CBR

The utility problem is observable in many learning systems including case-based reasoning (CBR). Indexing strategies have been implemented in CBR to overcome the effects of the utility problem but have been criticised as although improving retrieval efficiency they can reduce the competency of solutions and can be difficult to maintain. Here we present a novel indexing strategy based on a modified k-means clustering algorithm. We demonstrate that such an indexing strategy improves retrieval efficiency without adversely affecting solution competency. Importantly it also provides a means for the dynamic real time maintenance of retrieval knowledge thus ensuring that the index is always optimal.