Chapter 4 showed how to build a simple neural network for altering the ranking of search results based on what links users have clicked in the past. That neural network was able to learn which words in which combinations were important, and also which words were unimportant to a particular query. Neural networks can be applied to both classification and numerical prediction problems.
The neural network in Chapter 4 was used as a classifier—it gave a number for every link, predicting that the link with the highest number would be the one that the user would click. Because it gave numbers for every link, you could use all the numbers to change the rankings of the search results.
There are many different kinds of neural networks. The one covered in this book is known as a multilayer perceptron network, so named because it has a layer of input neurons that feed into one or more layers of hidden neurons. The basic structure is shown in Figure 12-3.
Figure 12-3. Basic neural network structure
This network has two layers of neurons. The layers of neurons are connected to each other by synapses, which each have an associated weight. The outputs from one set of neurons are fed to the next layer through the synapses. The higher the weight of a synapse leading from one neuron to the next, the more influence it will have on the output of that neuron.
As a simple example, consider again ...