To deal with a problem of incomplete comparisons in decision maker preference assessment, a Sonia (self-Organized network inspired by Immune Algorithm)-based decision neural network (DNN) is proposed the mutation mechanism of Sonia deals with a limited number of training data resulting from incomplete pair-wise comparisons by decision maker. Numerical ekperimets on Lp-metric function as underlying decision maker preference show that the errors of SONIA-based DNN are 1/4 times lower than those of conventional DNN for decision maker preference assesment with incomplete comparisons. Experiments on dish-up and fire-work data of a chain restaurant work assignment problem show the applicability of the proposed method to a real-world restaurant work manager preference modeling. Moreover, a multi-agent framework of an integrated restaurant work scheduling system is also proposed.