Modeling the networked software interactive behavior is the basis of understanding its internal mechanism and the running rules .The software interactive behavior log files are firstly collected by monitoring, and then the invariant constraints are mined from it, finally, a dynamic model learned from the finite state machine is presented. In this model, not only thought over the situation that the parallel partly ordered event sequence generated by the networked software interaction, but the interplay between data values and components interactions are also considered, and the event are satisfied the invariant constraints. In order to ensure certainty and completeness of the model, the method of merging equivalent states in the divided sub-diagram is proposed. The corresponding algorithms are also designed. Meanwhile, the effectiveness and feasibility of the proposed method are validated through experiments.