Is there a way to reset Recommendations?


#1

I used to love using the Recommended page to think about which films I should try watching soon. Having used this for a while, I found there was a ‘Watchlist’ feature. I then went through adding these recommended films to the Watchlist, as well as many other films that I also had not yet watched. A few days later, I returned to the Recommended page to see that it had completely changed, and consisted of a bunch of films I didn’t really care for, and was a little surprised they were now recommended to me. At first, I assumed it was just due to updated algorithms or whatever, but then I realised that none of the new films in the Recommended list were in my Watchlist. Therefore, adding films to the Watchlist removed them from my Recommended list.

I then went back to my Watchlist and cleared all 100+ films in the hope that the original recommended films would return to the Recommended page. Unfortunately, this has not happened after several weeks. Is there any way to reset this? It is one of my favourite features, and wish to use it again.

Thanks in advance!


#2

I don’t know how to reset the recommendations on iCM, but I do have an alternate list of recommendations for you, based on your favorites, but using a different algorithm. I’d be interested in hearing if you think these are better or worse than the recommendations on iCM.

Gladiator
Sin City
The Prestige
Fast Five
Interstellar
The Usual Suspects
American History X
Snatch
El laberinto del fauno
Intouchables
Reservoir Dogs
Tucker and Dale vs Evil
Memento
The Departed
Ghostbusters
Edge of Tomorrow
Cidade de Deus
American Beauty
Children of Men
Platoon


#3

Hello. Thanks for the response.

I think some of those films were on my previous list of Recommended films (at least Gladiator, Sin City, Interstellar, Memento, The Departed), so the algorithm seems to be pretty good. How did you formulate this list?


#4
  1. Download the favorites of the top 500 iCM users (or at least, the ones that I can access).
  2. For each film, calculate an “unweighted vote count” by counting the number of occurrences.
  3. Calculate a similarity score for each user, based on how much their list overlaps with yours, where score = overlap / sqrt(list size).
  4. For each film, calculate a “weighted vote count”, where users with a higher score have more weight.
  5. For each film, calculate score = (weighted vote count) / (unweighted voted count + 2)^X, where X is a parameter that controls how obscure the recommendations are (I can’t remember if I used 0.8 or 0.85). Then sort the films by score.

#5

Interesting. Did you do that in Excel, or use code/script-based software such as R?


#6

I used Java.


#7

I was wondering whether this should be moved to the Bugs board, as I think that it probably is an unintentional error.