Short-term voltage stability (STVS) is closely related to load dynamics. Recently, load models have been periodically tracked through ambient signal-based identification, through which the role of load dynamics in STVS can be elaborated in more detail. In this paper, a load dynamic stability index (LDSI) is proposed, which can be applied in STVS assessment and control. To our knowledge, this is the first time that online periodically identified load models are directly applied in analyzing the role of load dynamics in STVS. According to the periodical load modeling results, the dynamic process of a load during a voltage dip event is analyzed, after which the basic form of the LDSI is proposed. Based on the support vector machine method, a data-driven method is proposed to modify the LDSI to improve its properties. The modification is performed based on the simulation results in the WSCC 9 bus system to obtain the final form of the index, and the sufficiency and monotonicity of the index are validated under various operating conditions. Finally, the practical application of the proposed index to improve the STVS of a practical power system is discussed and illustrated in the IEEE 39 bus system.