Semi-supervised domain adaptation (SSDA) seeks to improve model generalization performance when there is only a small amount of data for which labeling information exists for the target domain data to be predicted. However, while recent SSDA models generalize well enough for discrepancies between source and target domain, they do not do a remarkable job of resolving intra-domain discrepancies that occur within unlabeled target domains. To address this, we propose a novel Curriculum and Contrastive learning-based Semi-supervised Domain Adaptation method, C2SDA. We use curriculum learning, which uses prediction entropy as a criterion for training difficulty and prioritizes learning easy unlabeled target domain data to achieve generalization performance. Furthermore, we apply contrastive learning, which enables cross-domain and cross-class discriminative feature learning, to further improve the generalization performance of the model. Experimental results demonstrate that our proposed method outperforms other SSDA methodologies in prediction by superiorly reducing inter- and intra-domain discrepancies.