About the order of multi label

How do I know according order of the origin corpus, for example
turn on the air conditioner and then raise the temperature

there are two intents in order.
But predictions from model is in no order
INDArray predictions = net.outputSingle(input);
So how can I get the right prediction with the same order of original corpus

If you have a sequence dependent order, it might be more reasonable to use a recurrent architecture with only one result each.

If you want to keep using a multi label system, you always get those results simultaneously, so you will have to create a set of rules of how to order them.

If I switch from TextCNN to bi-lstm,how to do multi label classification?
Is there any examples ?

There aren’t any examples that I’m aware of. But the idea is pretty simple: Instead of having an output of [0, 1, 1, 0, 1], you get a sequence of outputs [0, 1, 0, 0, 0], [0, 0, 1, 0, 0], [0,0,0,0,1]. Effectively making it a sequence of the labels - ideally it should be in the same order that you want to get out.

But, again if the multi label version works fine for you otherwise, having a rule based system for what actions to do in which order may be easier to make it work.

thank you.
I will consider each implementation