4 ms·

You are way off here. The article was not a customer support complaint about his personal account. Sure he could've written this article a helluva lot better,

You are way off here. The article was not a customer support complaint about his personal account. Sure he could've written this article a helluva lot better, but his point is 100% valid which you implicitly agree with by recommending he open multiple profiles. Remember that this guy is a writer not an engineer. His solution is likely bad! So throw out the idea of basing ratings on times of day or mood, etc. The problem statement is still there and valid: My preferences change over time. How can a good recommendation algorithm account for that? Brainstorm away...

frossie 17y ago
My preferences change over time. How can a good recommendation algorithm account for that? I'm confused. You just re-rate the discrepancies - I do that routinely (some movies seem okay after you watch them, then a few days later you realise they were eminently forgettable). The Netflix UI makes this easy by showing you the movies you recently rated for a while afterwards. I'd be pretty surprised if the recommendation algorithm didn't have a time weighting in it too, counting the movies you have rated recently more than the ones you have rated in the distant past. You also seem to think that multiple profiles is a workaround - well to me, it's a really good solution to that kind of problem. When Netflix announced they were going to pull the feature I was one of the many people who complained, and was very happy when they reversed themselves. "Remember that this guy is a writer not an engineer. His solution is likely bad! [..] The problem statement is still there" But is it? I don't have any personal evidence that it is, and dude says it is but he doesn't rate movies so how can I know? Is it for you? Do you find that Netflix recommends movies you would have liked 5 years ago but no longer do so because your tastes have changed? If so then yes, we have a problem. I do have a problem with the Netflix recommendation algorithm in that I don't feel it deals well enough with klunkers - just because I enjoyed 5 Hugh Jackman movies does not mean I will enjoy Van Hesling, because Van Helsing was a really bad movie and would not have been a good movie even with the second coming of James Dean. The solution to that is pretty simple technically though.
brg 17y ago
I don't think the gp is far off. The author of the article clearly says that he places movies in his queue which he does not watch. If he provided truthful responses to Netflix (not interested, not-enough-stars-to-order-again) regarding these then the algorithm could be given a chance. But he doesn't use the algorithm, and instead chooses movies the he won't watch. And then complains that Netflix will not populate his recommendations with more movies he doesn't want to watch. Non-linear functions (such as mood over time) and non-static behavior (attitude adjustment to one title due to exposure to another) can be accounted for in the Netflix recommendation algorithm.
- kevinpet 17y ago
  Precisely -- his opinion about netflix is based off misusing it. If he provided reasonable ratings, only ordered the types of movies he wants to watch, and then said "it isn't working because my moods change", then I'd pay a little more attention.
- jpwagner 17y ago
  Ok, the author is a bit of a moron. He said stupid things and detracted from what I am guessing his point is. He also didn't read about the Netflix algorithms and how they actually work (all around terrible job of reporting indeed.) However, to circle back, what I think is his point is interesting: how do you model "mood over time" and "non-static behavior". You say they "can be accounted for" and don't talk about how. In the winner's documentation they say they treated "user preference" variation over time the same as chunked up "user bias" over time (they define these terms in the paper.) While that obviously performed well numerically, isn't it believable that this is not an ideal solution, but one done under severe time constraints? The problem is by no means solved...that's all I'm saying.
  - roundsquare 17y ago
    If we just take his examples though, then the problem is on its way to being solved. So he puts movies on his queue that he doesn't watch? Great, he should give them low rankings. If the algorithm is good, that problem is solved. He likes some movies at night, some on the weekends and some with his wife? Well, okay, rate the movies how you thought they were when you saw them. The algorithm will recommend some movies for each situation. Your preferences change over time? Thats a bit harder, but was also out of the scope of the challenge. They were given static data. If you want to account for changes over time, then give the ratings some weighting that decreases over time or only use the last x months of data. But, like you said, he probably doesn't know much about the algorithm.
pmjoyce 17y ago
"I haven't devoted nearly enough time to figuring out the difference between four stars and five stars" I grappled with a similar issue a few months ago when I wrote about the 5 star rating system used by Yelp. If time and mood are factors in evaluating a movie at a point in time (when the movie itself doesn't change over time) then the issue is muddled further when using a 5-star rating system for bar or restaurants where the same person can have wildly different empirical experiences in the same day. It seems to me that it's almost impossible to capture all the variables necessary to accurately capture and assess how you feel about something at a given point in time (movie/restaurant/whatever). Even if we could, would enough people want to go though the doubtless cumbersome and time consuming process of collection enough of the time to draw any meaningful conclusions?

frossie 17y ago

My preferences change over time. How can a good recommendation algorithm account for that? I'm confused. You just re-rate the discrepancies - I do that routinely (some movies seem okay after you watch them, then a few days later you realise they were eminently forgettable). The Netflix UI makes this easy by showing you the movies you recently rated for a while afterwards. I'd be pretty surprised if the recommendation algorithm didn't have a time weighting in it too, counting the movies you have rated recently more than the ones you have rated in the distant past. You also seem to think that multiple profiles is a workaround - well to me, it's a really good solution to that kind of problem. When Netflix announced they were going to pull the feature I was one of the many people who complained, and was very happy when they reversed themselves. "Remember that this guy is a writer not an engineer. His solution is likely bad! [..] The problem statement is still there" But is it? I don't have any personal evidence that it is, and dude says it is but he doesn't rate movies so how can I know? Is it for you? Do you find that Netflix recommends movies you would have liked 5 years ago but no longer do so because your tastes have changed? If so then yes, we have a problem. I do have a problem with the Netflix recommendation algorithm in that I don't feel it deals well enough with klunkers - just because I enjoyed 5 Hugh Jackman movies does not mean I will enjoy Van Hesling, because Van Helsing was a really bad movie and would not have been a good movie even with the second coming of James Dean. The solution to that is pretty simple technically though.

brg 17y ago

I don't think the gp is far off. The author of the article clearly says that he places movies in his queue which he does not watch. If he provided truthful responses to Netflix (not interested, not-enough-stars-to-order-again) regarding these then the algorithm could be given a chance. But he doesn't use the algorithm, and instead chooses movies the he won't watch. And then complains that Netflix will not populate his recommendations with more movies he doesn't want to watch. Non-linear functions (such as mood over time) and non-static behavior (attitude adjustment to one title due to exposure to another) can be accounted for in the Netflix recommendation algorithm.

kevinpet 17y ago
Precisely -- his opinion about netflix is based off misusing it. If he provided reasonable ratings, only ordered the types of movies he wants to watch, and then said "it isn't working because my moods change", then I'd pay a little more attention.
jpwagner 17y ago
Ok, the author is a bit of a moron. He said stupid things and detracted from what I am guessing his point is. He also didn't read about the Netflix algorithms and how they actually work (all around terrible job of reporting indeed.) However, to circle back, what I think is his point is interesting: how do you model "mood over time" and "non-static behavior". You say they "can be accounted for" and don't talk about how. In the winner's documentation they say they treated "user preference" variation over time the same as chunked up "user bias" over time (they define these terms in the paper.) While that obviously performed well numerically, isn't it believable that this is not an ideal solution, but one done under severe time constraints? The problem is by no means solved...that's all I'm saying.
- roundsquare 17y ago
  If we just take his examples though, then the problem is on its way to being solved. So he puts movies on his queue that he doesn't watch? Great, he should give them low rankings. If the algorithm is good, that problem is solved. He likes some movies at night, some on the weekends and some with his wife? Well, okay, rate the movies how you thought they were when you saw them. The algorithm will recommend some movies for each situation. Your preferences change over time? Thats a bit harder, but was also out of the scope of the challenge. They were given static data. If you want to account for changes over time, then give the ratings some weighting that decreases over time or only use the last x months of data. But, like you said, he probably doesn't know much about the algorithm.

pmjoyce 17y ago

"I haven't devoted nearly enough time to figuring out the difference between four stars and five stars" I grappled with a similar issue a few months ago when I wrote about the 5 star rating system used by Yelp. If time and mood are factors in evaluating a movie at a point in time (when the movie itself doesn't change over time) then the issue is muddled further when using a 5-star rating system for bar or restaurants where the same person can have wildly different empirical experiences in the same day. It seems to me that it's almost impossible to capture all the variables necessary to accurately capture and assess how you feel about something at a given point in time (movie/restaurant/whatever). Even if we could, would enough people want to go though the doubtless cumbersome and time consuming process of collection enough of the time to draw any meaningful conclusions?