2
$\begingroup$

I would like to create a troubleshooting wizard.

The user will go through the wizard and choose different options, what options they choose will determine what is displayed next in the wizard.

Eventually the user will solve their problem (or not), at the end they will choose 'yes' or 'no' if there problem was solved.

The wizard would learn what most common resolutions to a problem was based on past input.

Before I go exploring machine learning further, does this sound like the right tool for the job?

$\endgroup$
1
$\begingroup$

There is lack for details here, but problem is very interesting, sounds like QA chat-bot (you may read more about these) a bit. My thoughts for this solution to be viable:

  1. there should be sufficient amount of users of your wizard to collect necessary amount of data if you want to use deep learning here. If you only have several hundred of cases when wizard is used, forget it, use some simple statistical analysis (like linear regression, naive Bayes etc.).
  2. For collected data being variable enough to cover different cases, there should be some randomization in collecting the data - sometimes wizard should be asking questions in different order and maybe even asking irrelevant questions.
  3. As for architecture, you may want to classify sequences of questions/answers, then the most obvious choice is variant of RNN. Or maybe you want to classify how good the next question is - maybe you want to use decision tree/random forest here to classify it. Or maybe you need reinforcement learning? That was to point you should think very well what you want to classify and then test it, maybe in several iterations.

So, if the wizard is complex enough and amount of data you have or may collect is big, then answer is most probably yes.

$\endgroup$
2
$\begingroup$

Yes and no, it sounds ok, but depends on the complexity of the wizard very much. It is possible that a simple rule-based algorithm will be super useful and easier to implement.

If you really want to try it (for example for learning purposes), the bottleneck could be the data which needs to be available for initially training the model. You should have enough user queries (or at least simulated queries) even before your ML wizard exists so that the model can learn.

If it is something you need for work, I would consider consulting somebody with ML/Data Science experience with the specific details of what you need.

Hope, this helps at least a bit.

Best regards, Jakub

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.