Text-based emotion analysis, an important task in Natural Language Processing (NLP), aims to identify and understand emotional tendencies in text. Recently, given their strong performance in text classification, Graph Neural Networks (GNNs) have been utilized in various emotion recognition studies. They have excellent structural modeling abilities but lack context encoding strength. On the other hand, Large Language Models (LLMs) such as BERT and GPT are specially designed to model the text context. Aiming to utilize both their advantages, we investigated several ways to combine GNNs with LLMs for the emotion recognition task. First, we used BERT to generate embeddings for the graph document nodes. Next, we extended the system to include a description of the input data’s emotional content obtained from GPT as an additional node embedding. For experiments and system evaluation, we used the GoEmotions dataset. The results clearly show that combining GNN and LLM improves the emotion classification performance by 20% to 30% compared to when either GNN or LLM is used alone.