I am using keras MLP network for binary classification of 3-D word vector input_shape=(None,24,73). I have used two dense layers dense_1 and
input_shape=(None,24,73)
dense_1
Since you have a binary_classification task your last layer should look something like this
model.add(Dense(1, activation='sigmoid'))
Right now you model is out puting 3D array which don't match the shape of your target (2D)