Page 153 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

142 Chapter 5 Depression discovery in cancer communities using deep learning

4. Models

We describe the four deep neural network models, which can
be used in cancer communities for depression detection. The first
model discussed is convolutional neural network (CNN) and its
variants that can be used for the performance enhancements;
the other one is RNN. Both the models can be built on word
embedding described in the previous section.

4.1 Convolutional neural network
A CNN is feed-forward neural network where the layers do not
form the cycle to give the feedback back to the model to train it-
self. It is a perceptron model with multiple layers. The CNN
model would like as shown in Fig. 5.5.
In a standard CNN model, the first layer of model is the input
layer that uses an embedding word matrix where each sentence is
converted into the word matrix using the word embedding model
described in the previous section. This word embedding is low-
dimensional representation, but it could also be one-hot vector
that indexes the word into a vocabulary. Let w i ˛V k be the k-
dimensional word vector corresponding to the i-th word in the
sentence or can be said as grouping parameter for words. This k
allows the model to work on different resolution of word vectors
such as 2-grams, 4-grams, and so on. In word vector format, a sen-
tence of length n (padded where necessary) can be represented as:
w 1:n ¼ w1 4w2 4w3 4/4wn; (5.1)

Representation of sentence into n*k Filtering Layer Max-Over Time Polling Soft-Max Output Layer
Figure 5.5 Basic structure of CNN feed-forward model. CNN, convolutional neural network.