Nonlinear nearest-neighbour matching and its application in legal precedent retrieval

Case-based reasoning (CBR) has been widely and successfully applied in legal precedent retrieval. Traditional nearest-neighbour (NN) matching has shown that it is not capable of dealing with the situations that the values of weights or dimensional matching scores are extremely high or low. These extreme situations have nonlinear psychological effects on the aggregate marching scores. Generalized nearest-neighbour (GNN) matching improved NN matching in certain situations, but it is not generally applicable and it can cause an unexpected ranking. In order to improve the limitation of NN matching and complement the deficiency of GNN matching, we propose a novel nonlinear nearest-neighbour (NNN) matching function based on the adjustments for nonlinear effects and the fuzzy logic inference. In this paper, we also describe how we apply NNN matching in our legal precedent retrieval system.