In a range of scientific coauthorship networks, transitions emerge in degree distributions, correlations between degrees and local clustering coefficients, etc. The existence of those transitions could be regarded as a result of the diversity in collaboration behaviours of scientific fields. A growing geometric hypergraph built on a cluster of concentric circles is proposed to model two specific collaboration behaviours, namely the behaviour of leaders and that of other members in research teams. The model successfully predicts the transitions, as well as many common features of coauthorship networks. Particulary, it realizes a process of deriving the complex "scale-free" property from the simple "yes/no" experiments. Moreover, it gives a reasonable explanation for the emergence of transitions with the difference of collaboration behaviours between leaders and other members. The difference emerges in the evolution of research teams, which synthetically addresses several specific factors of generating collaborations, namely the communications between research teams, the academic impacts and homophily of authors.