Spoken language comprehension is crucial for dialogue systems, involving two critical duties: intent recognition and slot filling. These obligations extract and differentiate the intention and entity specifics from user input. Notably, recent research has highlighted a significant correlation between these two duties, suggesting that a coordinated approach to their instruction could result in better outcomes. This article uses graph neural networks to present a joint model for recognizing intent and filling slots. The encoding layer employs the BERT pre-trained language model to support the semantic representation of input text. Additionally, it incorporates the graph attention mechanism to enhance the model's ability to capture semantic connections present within sentences. Empirical analyses support that the unified model achieves exceptional accuracy in both intent recognition and slot-filling areas. Specifically, it attains intent recognition accuracy rates of 98.4% and 99.1% and slot-filling F1 scores of 98.5% and 97.2%. Additionally, it achieves sentence-level semantic frame accuracy rates of 93.2% and 92.8% on the ATIS and SNIPS datasets, respectively. It is important to note that it outperforms other benchmark models.