So, I'm just starting to learn how a neural network can operate to recognize patterns and categorize inputs, and I've seen how an artificial neural network can parse image data and categorize the images (demo with convnetjs), and the key there is to downsample the image and each pixel stimulates one input neuron into the network.
However, I'm trying to wrap my head around if this is possible to be done with string inputs? The use-case I've got is a "recommendation engine" for movies a user has watched. Movies have lots of string data (title, plot, tags), and I could imagine "downsampling" the text down to a few key words that describe that movie, but even if I parse out the top five words that describe this movie, I think I'd need input neurons for every english word in order to compare a set of movies? I could limit the input neurons just to the words used in the set, but then could it grow/learn by adding new movies (user watches a new movie, with new words)? Most of the libraries I've seen don't allow adding new neurons after the system has been trained?
Is there a standard way to map string/word/character data to inputs into a neural network? Or is a neural network really not the right tool for the job of parsing string data like this (what's a better tool for pattern-matching in string data)?