All the machine learning algorithms or architectures that we have used so far require the input to be real-valued or matrices of real-valued quantities, and that's a common theme in machine learning. For example, in the convolution neural network, we had to feed raw pixel values of images as model inputs. In this part, we are dealing with text, so we need to encode our text somehow and produce real-valued quantities that can be fed to a machine learning algorithm. In order to encode input text as real-valued quantities, we need to use an intermediate science called Natural Language Processing (NLP).
We mentioned that in this kind of pipeline, where we feed text to a machine learning model such as sentiment analysis, this will be problematic and won't work because we won't be able to apply backpropagation or any other operations...
