The volume of vehicular network traffic is very context (time and geographic location) and technology-dependent. Considering both multi-hop geocast and single-hop broadcast techniques, the route availability can be affected by transient and permanent traffic variations. Therefore, our research tackles one of the most pressing challenges in vehicular ad-hoc networks (VANETs), i.e., accommodating fine-grained spatio-temporal variance in vehicular density over time and space. This article proposes a new framework called SpTFrame to achieve fast message dissemination. The proposed approach uses a software-defined vehicular networks (SDVNs) architecture along with a deep reinforcement learning (DRL) model. SpTFrame employs a convolutional neural network (CNN) and a gated recurrent unit (GRU) to detect spatio-temporal correlation under vehicle distribution on urban road networks. The novelty of the work is that it tackles short-term spatio-temporal volatility in SDVNs' inherent characteristics and offers a way to handle short-term network topology changes. The experimental results were obtained using real-world traffic data from Jodhpur, India, and open-source road network data from OpenStreetMap. The study results show that the proposed method improves efficiency and the network's performance. The framework can be useful in vehicular applications that require fast message dissemination.