Computational argumentation has emerged as a crucial domain within artificial intelligence, offering insights into reasoning, decision-making, and communication processes. Its applications span various real-world scenarios, from legal argumentation to intelligence analysis. In this framework, arguments are represented as nodes in a graph, with edges depicting the relationships (support or attack) between them. This study explores the capacity of neural network approaches to learn bipolar gradual argumentation semantics, which incorporates both supportive and oppositional relationships. We initiate our approach by deriving acceptability degrees for graph nodes using the Quantitative Argumentation Debate (QuAD) semantics. We then apply this method to two benchmark datasets: Twelve Angry Men and Debatepedia. Utilizing this data, we train and assess three neural network architectures: Multilayer Perceptron (MLP), Graph Convolution Network (GCN), and Graph Attention Network (GAT), evaluating their ability to learn the acceptability degrees generated by QuAD semantics. Our findings demonstrate that these neural network methods can learn bipolar gradual argumentation semantics. Notably, models based on the GCN architecture outperform the other two, highlighting the significance of explicitly modeling argumentation graphs. Our software is publicly available at: https://github.com/clulab/icaart24-argumentation .