# Verifying NLP prediction

In NLP multi-label classification

How can we verify multi-label classification is right?

I mean I want to highlight the words Machine used to give me the results I got.

an example on technical support labels

When you use a Dell mouse to click or select an icon, window, folder or other object in Windows the computer produces a clicking or other similar sound that you can hear through your speakers. If the clicking sound produced by the computer disturbs you or others in the office, disable the sound in Windows and enjoy a quieter and more productive computing experience. Click “Start,” then “Control Panel.” In the “Adjust your computer’s settings” window, click “Hardware and Sound.” Click “Change System Sounds” underneath the Sound link in the "Hardware and Sound" window to open the Sounds window. Click the “Sounds” tab. Scroll down to and select the “Start Navigation” option in the Windows Explorer section of the Program Events list box. Click the drop-down list labeled “Windows Navigation Start” and select “None.” Click “Apply,” then the “OK” button. Close the "Hardware and Sound" window. Tip You can also open the Sounds window by right-clicking the speaker icon in the taskbar and then clicking “Sounds” on the pop-up menu.

predictive model can give these labels

Operating System Problem

Sound Problem

Hardware Problem

I want to know what are the words the influenced machine to choose these labels.

If the text is so big, I cant read it all to verify if the labels were correct or not.

I want the machine to highligh/select/ or any other way tell me what made it choose these labels

Something like this would be good

When you use a Dell mouse to click or select an icon, window, folder or other object in Windows the computer produces a clicking or other similar sound that you can hear through your speakers. If the clicking sound produced by the computer disturbs you or others in the office, disable the sound in Windows and enjoy a quieter and more productive computing experience. Click “Start,” then “Control Panel.” In the “Adjust your computer’s settings” window, click “Hardware and Sound.” Click “Change System Sounds” underneath the Sound link in the "Hardware and Sound" window to open the Sounds window. Click the “Sounds” tab. Scroll down to and select the “Start Navigation” option in the Windows Explorer section of the Program Events list box. Click the drop-down list labeled “Windows Navigation Start” and select “None.” Click “Apply,” then the “OK” button. Close the "Hardware and Sound" window. Tip You can also open the Sounds window by right-clicking the speaker icon in the taskbar and then clicking “Sounds” on the pop-up menu.

• Determining "what a model is looking at" is an entire field of research, I don't think it can be succinctly answered here. See for example this article about how a neural network can learn the relevant parts of an image. – Frans Rodenburg Oct 10 '19 at 5:46
• Try searching for something along the lines of: "visualizing network weights" – Frans Rodenburg Oct 10 '19 at 5:47
• @FransRodenburg Thanks heaps – asmgx Oct 10 '19 at 7:42
• Not all models are black boxes or transparent the same way. Decision trees for instance, are famous for being quite interpretable, while neural networks are a lot harder to interpret. Coul you tell us in the question what models are you using in the classification task? I see the other comments talk about neural networks. It's that what you are using? – Armando Oct 10 '19 at 20:28
• @armando I see now that it was not explicitly stated in the question, but I assumed it because it seems to be the dominating strategy to NLP. – Frans Rodenburg Oct 11 '19 at 4:53