0
$\begingroup$

Are there data clustering algorithms which by definition require human judgement as part of the algorithm? What I mean is that once in a while the algorithm presents the human operator (or any other species of operator) with a query of whether some objects (not necessarily from the original data set) should be grouped into the same cluster. The human operator serves as an oracle for the algorithm, and drives it towards a meaningful clustering.

It seems that such algorithms should exists, because different clustering algorithms usually optimize different target functions which depends on parameters, and on the one hand the user should choose an algorithm and parameters based on her goal, but on the other hand it is common with clustering problems that the user doesn't know in advance what he is looking for.

I tried Google searching for such algorithms, using key words that I thought might be related (for example, "human in the loop", "oracle", "agent", "semi supervised", "human judgment"), but so far I haven't found anything.

$\endgroup$
  • 1
    $\begingroup$ Well supervised learning, such as GLM, is kind of what you are describing - except that the user intervene at the very beginning of the process (the clustering the algorithm is starting with) - $\endgroup$ – Riff Feb 9 '17 at 13:03
  • $\begingroup$ Reinforcement_learning uses repetead feedback to improve the result. Although the feedback does not neccesarily come from a human, it possibly can. en.wikipedia.org/wiki/Reinforcement_learning $\endgroup$ – Nikolas Rieble Feb 9 '17 at 13:05
  • 1
    $\begingroup$ @Riff What I'm looking for is an algorithm that decides what questions to ask the oracle, where on the one hand it tries to ask a small amount of questions, an on the other it tries to get the most information from the agent. Thus for example it may decide to ask a question about data which it generated (and does not belong to the data set) since an answer might provide more information. And also the answers to previous questions may influence which questions the algorithm asks next. $\endgroup$ – Lior Feb 9 '17 at 14:03
  • $\begingroup$ Then your Google is broken. google.com/search?q=semi-supervised+clustering $\endgroup$ – Anony-Mousse Feb 9 '17 at 22:33
  • 1
    $\begingroup$ @Anony-Mousse, All you had to say is that these are called active algorithms. $\endgroup$ – Lior Feb 10 '17 at 10:11

Your Answer

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

Browse other questions tagged or ask your own question.