This paper introduces a Dynamic Convolutional Neural Network (DCNN) for semantic sentence modeling. The DCNN uses Dynamic k-Max Pooling, a global pooling operation over linear sequences, to capture both short and long-range relations in sentences. The network handles varying sentence lengths and induces a feature graph that captures semantic relationships without relying on parse trees. It is applicable to any language and achieves excellent performance in four tasks: binary and multi-class sentiment prediction, six-way question classification, and Twitter sentiment prediction using distant supervision. The DCNN outperforms other models in the first three tasks and achieves a greater than 25% error reduction in the last task compared to the strongest baseline. The network's architecture includes wide convolutional layers and dynamic k-max pooling layers, which allow for the extraction of higher-order and longer-range features. The DCNN is sensitive to word order and captures both local and global semantic relations. It is trained using backpropagation and Adagrad optimization, and it performs well on tasks involving sentiment analysis, question classification, and Twitter sentiment prediction. The network's feature detectors are able to recognize patterns within n-grams and capture syntactic, semantic, or structural significance. The DCNN is a flexible and effective model for sentence modeling that can be applied to various tasks without relying on external resources or parse trees.This paper introduces a Dynamic Convolutional Neural Network (DCNN) for semantic sentence modeling. The DCNN uses Dynamic k-Max Pooling, a global pooling operation over linear sequences, to capture both short and long-range relations in sentences. The network handles varying sentence lengths and induces a feature graph that captures semantic relationships without relying on parse trees. It is applicable to any language and achieves excellent performance in four tasks: binary and multi-class sentiment prediction, six-way question classification, and Twitter sentiment prediction using distant supervision. The DCNN outperforms other models in the first three tasks and achieves a greater than 25% error reduction in the last task compared to the strongest baseline. The network's architecture includes wide convolutional layers and dynamic k-max pooling layers, which allow for the extraction of higher-order and longer-range features. The DCNN is sensitive to word order and captures both local and global semantic relations. It is trained using backpropagation and Adagrad optimization, and it performs well on tasks involving sentiment analysis, question classification, and Twitter sentiment prediction. The network's feature detectors are able to recognize patterns within n-grams and capture syntactic, semantic, or structural significance. The DCNN is a flexible and effective model for sentence modeling that can be applied to various tasks without relying on external resources or parse trees.