This paper presents a deep convolutional neural network (CharSCNN) designed to perform sentiment analysis on short texts, such as single sentences and Twitter messages. The network leverages character-level to sentence-level information to capture contextual details, which are often limited in such texts. CharSCNN uses two convolutional layers to extract relevant features from words and sentences of any size, utilizing word embeddings from unsupervised pre-training. The network is evaluated on two corpora: the Stanford Sentiment Treebank (SSTb) for movie reviews and the Stanford Twitter Sentiment corpus (STS) for Twitter messages. For SSTb, CharSCNN achieves state-of-the-art results with 85.7% accuracy in binary positive/negative classification and 48.3% accuracy in fine-grained classification. For STS, it achieves 86.4% accuracy. The paper also discusses the effectiveness of unsupervised pre-training, character-level features, and sentence-level features in detecting negation. The results demonstrate the robustness and effectiveness of CharSCNN in sentiment analysis tasks.This paper presents a deep convolutional neural network (CharSCNN) designed to perform sentiment analysis on short texts, such as single sentences and Twitter messages. The network leverages character-level to sentence-level information to capture contextual details, which are often limited in such texts. CharSCNN uses two convolutional layers to extract relevant features from words and sentences of any size, utilizing word embeddings from unsupervised pre-training. The network is evaluated on two corpora: the Stanford Sentiment Treebank (SSTb) for movie reviews and the Stanford Twitter Sentiment corpus (STS) for Twitter messages. For SSTb, CharSCNN achieves state-of-the-art results with 85.7% accuracy in binary positive/negative classification and 48.3% accuracy in fine-grained classification. For STS, it achieves 86.4% accuracy. The paper also discusses the effectiveness of unsupervised pre-training, character-level features, and sentence-level features in detecting negation. The results demonstrate the robustness and effectiveness of CharSCNN in sentiment analysis tasks.