Abstract. On the one hand, for the limitation that a user in Interactive Evolutionary Computation (IEC) is apt to feel tired during the solutions evaluation, the result of IEC is not so desirable. On the other hand, the many-years and many-users accumulated data of the evaluation in IEC is enormous, which could be a pool of information about users' preference. By mining the big data of accumulated evaluation information, preference models can be established to form a model repository. When a new coming customer tries to get his/her favorite design aided by IEC, the nearest model in the repository can be selected as the result of model matching and the corresponding optimum can be directly transferred to accelerate the optimization. For this purpose, transfer learning is adopted. The simulated experiments validated this scheme and showed that the performance of big-data supported IEC is more related with volume of the models in big data, but less related with problems' difficulty.