Data-Driven Personalized Individual Semantic Space Learning Methods for Linguistic Multi-attribute Group Decision Making

Learning the personalized individual semantic (PIS) space of decision-makers based on historical data is a key step toward facilitating the effective implementation of linguistic multi-attribute group decision making (MAGDM). To this end, this study proposes a data-driven PIS space learning method under the linguistic MAGDM framework. Based on the historical data, a two-stage PIS space learning model is developed. The first-stage model aims to maximize the difference in comprehensive scores between alternatives from adjacent categories while minimizing the difference between alternatives within the same category. The second-stage model determines the weights of the sampled personalized numerical scales (PNS) obtained in the first stage by maximizing the consensus level among decision-makers. Regularization terms are incorporated into the objective functions of both stages of the convex quadratic programming models to prevent overfitting. In addition, linguistic distribution assessments (LDAs), as a powerful form of linguistic representation, are employed to express the preference information of decision-makers. Finally, a car evaluation case is used to demonstrate the application of the proposed method, and two comparative analyses—robustness and consensus level—are conducted to validate the effectiveness and superiority of the proposed method.