In this example, we will train a one-dimensional convolutional neural network (CNN) to classify sentences as either positive or negative. You have already seen how to classify images using two-dimensional CNNs in Chapter 3, Deep Learning with ConvNets. Recall that CNNs exploit spatial structure in images by enforcing local connectivity between neurons of adjacent layers.
Words in sentences exhibit linear structure in the same way as images exhibit spatial structure. Traditional (non-deep learning) NLP approaches to language modeling involve creating word n-grams (https://en.wikipedia.org/wiki/N-gram) to exploit this linear structure inherent among words. One-dimensional CNNs do something similar, learning convolution ...
