Recently, with the proliferation of social networks, reviews and user-item interactions have been widely used and recognized as valuable resources for enhancing representation learning in recommendation systems. On the other hand, deep learning techniques, particularly Graph Neural Networks (GNNs), are at the forefront of current research due to their ability to capture complex non-linear relationships. Nevertheless, few studies have explored the combination of sentiment analysis and GNNs within the context of social networks. In this paper, we propose SGCN-SRec, a novel Sentiment-based Graph Convolutional Network for Social Recommendation. This approach consists of three modules: (1) a sentiment analysis module using the pre-trained BERT model combined with CNN and LSTM models; (2) an interaction graph and social information module; and (3) a rating prediction module. The proposed structure generates embeddings through a GCN and combines them using an attention mechanism. Specifically, a user-item interaction graph, a friendship and a trust user-user relation graphs are considered. Experimental results demonstrate that our model surpasses the performance of current state-of-the-art methods.