Temporal networks, which are defined by their changing structures over time, present considerable difficulties for accurate link prediction. Conventional static models are inadequate for capturing dynamic interactions, thus reducing their effectiveness in forecasting future links. This study introduces an innovative deep learning framework that merges centrality-aware Graph Convolutional Networks (GCNs) for generating embeddings with a Long Short-Term Memory (LSTM) network for temporal classification. The GCN component incorporates node centrality metrics into the aggregation process, ensuring that the resulting embeddings represent the network’s structural and positional significance. These embeddings are calculated across various graph snapshots, creating time-series trajectories for each node. The LSTM model is subsequently trained to understand these embeddings’ temporal progression and predict the probability of future link formation between node pairs. Our method effectively captures dynamic link patterns by accounting for both temporal dependencies and structural changes. Experimental tests on standard temporal network datasets show that the proposed method surpasses traditional classifiers and static models, achieving higher accuracy and robustness in predicting future interactions.