In a filter bubble, what are some of the techniques and best ways to determine the relationship between a video that is watched or liked, and the list of related videos to be presented to the browsing user? Thanks.

  • 2
    $\begingroup$ It is not clear what do you expect. Do you want to reverse personalization algorithm based on links? Or you want to know which techniques are good? The first one is hard because you do not have access to the algorithm only the part of your data, and the second is rather broad question. Everyone uses own kind of personalization algorithm and try to maximize personalization based on some metric comparing users, movies, votes etc. There is nothing like the best one, also there are several independent experiments that show it does not work as expected. $\endgroup$ – Evil May 29 '16 at 20:31
  • $\begingroup$ Can you please quote such experiments? What do you mean by it doesn't work as expected. Are you saying all these personalization algorithms are somewhat flawed? $\endgroup$ – Jack Maddington May 29 '16 at 21:11
  • 1
    $\begingroup$ from TED about cutting information flow. Another ones are a bit hard - statistical inconsistency (I have to find link), anyway, the amount of data used is modeled from data they gather (the do not have access to latent indicators) so basically if you try to base your results on partial, inconsistent and not interchangeable data (people do differ, ok?), the results got huge margin of error, but it would make filters non-working at all, so all data is assumed to explain the whole variance of results - results get very biased.. $\endgroup$ – Evil May 29 '16 at 21:28
  • $\begingroup$ Anyway if you are interested in such filters for theoretical reasons or you can acquire more data, it might be working for you. So back to the point, it is not crystal clear what you are asking about. $\endgroup$ – Evil May 29 '16 at 21:29
  • 1
    $\begingroup$ Ok, so one side of view you know already, there is part about making estimators or the actual metric - this is very hard part, the whole concept works as far objective function, metric, similarity function works, so it is not that bad, only finding consistent functions is really hard... Let me just add that comparing people by behaviour (which is exactly what recorded data is) does not guarantee the good measure of similarity between them. $\endgroup$ – Evil May 29 '16 at 21:32

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.