Using Domain Knowledge to Boost Case-Based Diagnosis: An Experimental Study in a Domain with Very Poor Data Quality

The quality of case data can be an important factor for a Case Based Reasoning (CBR) system. In our research, we are facing a diagnostic problem in a product maintenance domain, where a large volume of low quality case data is collected. In this paper, we report an experimental study of whether using domain knowledge can improve the performance of either non-diversified or diversified retrieval in this domain. As each case can be associated with some domain knowledge that can be easily obtained in this domain, we try to use it to assist case matching. Our experimental results show that the domain knowledge can significantly improve the performance of both the non-diversified and the diversified approaches. Furthermore, using the diversified approach seems to be of greater value when also using domain knowledge.