1
$\begingroup$

I have data that looks like this:

priority    task         area 
1           clean room   living room
2           clean room   living room
2           water plants bedroom

I want to identify records with a priority of 2 that should be a priority of 1 based on similarity between the priority 2 task and each priority 1 task.

The result data would look like this:

priority    task         area         propensity_for_priority_1
1           clean room   living room  1
2           clean room   bedroom      .5
2           water plants bedroom      0

String similarity does not need to be performed (i.e. living room and bedroom are mutually exclusive). Is there a matching algorithm that can be used to compare various fields and result in one propensity_for_priority_1 score for each record?

$\endgroup$
2
  • $\begingroup$ Do you have a list of tasks labelled as priority 2 that you know should be priority 1? And/or should not be priority 1? $\endgroup$
    – Akavall
    Commented Mar 10, 2020 at 23:14
  • $\begingroup$ Looks like you are trying to compare how match the records are with those priority 1 records. And since you have only two field, "task" and "area", so I guess the propensity_for_priority_1 will be one of the values in {0, .5, 1}? $\endgroup$
    – 1tan Wang
    Commented Mar 11, 2020 at 5:21

1 Answer 1

0
$\begingroup$

2 Approaches.

a) Siamese networks, let the net tell you whats similiar and what lies together.

b) just perform multiclassification if the data permits, I dont know all the details but it looks like it.

$\endgroup$
1
  • $\begingroup$ yep - I'm using a random forest right now that's working pretty well. however, I was wondering if there are methods that are a notch below actual modeling that would be equally as effective. $\endgroup$ Commented Mar 11, 2020 at 12:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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