In multi-robot plume source searching with sporadic cues, the classic approaches strive for achieving social information consistency of all robots while the exploration ability of individual robot is ignored, which weakens the adaptivity of the group in complex environment. To overcome this drawback, a cooperative infotaxis searching approach is proposed. The relative entropy is introduced to measure the cognitive differences of likelihood function of source location between robots. Then, different weights are assigned to the sensor measurements of individual robot based on the cognitive differences. In the Bayesian learning process, the trade-off between individual cues and social cues is adaptively regulated for acquiring private source location probability distribution. Finally, the collaborative infotaxis search strategy is implemented by performing an entropy decision of each robot. The advantages of the proposed method are illustrated by simulation experiments under different scenarios.