Keras中具有不同数据类型的深度学习模型

Question

I trying to make a classification model with Keras. My data contains some numeric features and some text features. By text features I mean comments or something similar. Numeric features will be category, age so on.

I want to pass the text feature to a Embedding layer and then to a LSTM layer. The numeric feature needed to be passed to a series of Dense Layers. After that both layers needed to be concatenated. After that a Dense Layer to make the output.

How can I implement this type of model in Keras.?

Or is there any other way to use both numeric features and text based features in the model at the same time.?

Answer 1

It is fairly easy to implement such a network using keras functional API. suppose you have defined two sequential models to process your textual and numerical features, you can then merge the output and continue with more layers:

txt_input = keras.layers.Input(shape=(n,))
txt_feat = text_network(txt_input)
num_input = keras.layers.Input(shape=(m,))
num_feat = num_network(input2)
concatinated = keras.layers.Concatenate()([txt_feat, num_feat])

out = keras.layers.Dense(nodes)(concatinated)
model = keras.models.Model(inputs=[input1, input2], outputs=out)

you can also use other types of merge using any merge layer from keras.