Discovering Weight Knowledge Based on Linguistic Information Aggregation

CHEN Shuguo,WANG Dongfa,ZHU Haijun,DENG Mingrong,MA Hong

2021-01-01

Abstract:Weight knowledge is important in many fields in decision science. For example, in an artificial neural network (ANN), knowledge resides in the weights on links between neural neurons, based on which output is obtained from given inputs. Although a meaningful output of an ANN is of great use, implications of individual weights are often difficult to decipher. In the context of qualitative decision making, linguistic variables often express information that is about ordinal ranking rather than an exact quantity, and therefore the challenge lies in how to determine the weights conforming to the reality and making sense of the decision. In this work, we discriminate two kinds of weights, where one kind is related to the importance of attributes, and the other is related to aggregation characteristics. We propose a weight estimation method, in which these two kinds of weights are calculated simultaneously from a set of known cases that provide additional aggregation information. Using a simulated annealing algorithm, weights satisfying certain conditions can be obtained, and results for new cases can be aggregated thereafter. We find that the weights developed in our method are more explanatory than other approaches, such as the artificial neural network.