Sarcasm detection is an essential task that can help identify the actual sentiment in user-generated data, such as discussion forums or tweets. Sarcasm is a sophisticated form of linguistic expression because its surface meaning usually contradicts its inner, deeper meaning. Such incongruity is the essential component of sarcasm, however, it makes sarcasm detection quite a challenging task. In this paper, we propose a model which incorporates emotion and sentiment features to capture the incongruity intrinsic to sarcasm. Moreover, we use CNN and pre-trained Transformer to capture context features. Our approach achieved state-of-the-art results on four datasets from social networking platforms and online media.