User fatigue significantly restricts the practical applications of interactive genetic algorithms for complicated optimization problems. To alleviate user fatigue, an interactive genetic algorithm is presented in this study, where individuals are evaluated with varying accuracy. In the developed algorithm, multiple language sets with different granularities are proposed and applied to evaluate individuals. A subset of the whole language sets is chosen first to evaluate a population, adaptively to the convergence of the current population. For an individual in the current population, an appropriate language set is chosen from the subset to evaluate it according to the distance between the individual and the user's preferred region. The proposed algorithm is compared with some other algorithms in literature on curtain design. Empirical results demonstrate that the developed algorithm can significantly alleviate user fatigue and improve search efficiency.