Hi all. I'm not sure if criticker's algorithm already works like this or if it's been suggested already, but I think that the algorithm should take into account the fact that not all users have seen the same films and only those films.
To do this, I propose that when calculating the TCI of a user, only the films that both users have seen are used in the equation. So instead of calculating the percentile scores of films based on all the films seen by a user, the percentiles should be based only on all the films both users have in common.
To give an example scenario, imagine there are two users:
User 1
Films watched - Score
Film A - 50
Film B - 0
Film C - 100
User 2
Films watched - Score
Film A - 100
Film B - 50
Film D - 0
Using my suggested system to calculate the TCI of User 2 for User 1, only Films A and B will be used which for both users will have percentile scores of 100 and 0 respectively in that case.
Take into account the fact that not all users have seen the same films only
- uvlalid
- Posts: 8
- 1339 Ratings
- Your TCI: na
- Joined: Tue May 15, 2018 6:21 pm
- mpowell
- Posts: 3885
- 1201 Ratings
- Your TCI: na
- Joined: Fri Sep 09, 2005 10:22 am
Re: Take into account the fact that not all users have seen the same films only
Hi! Yes, actually we do this already. Only the films in common are considered when building the TCI
You can read more about that here: https://www.criticker.com/explain/
Thanks for joining the forum!!
You can read more about that here: https://www.criticker.com/explain/
Thanks for joining the forum!!
- livelove
- Posts: 759
- 67 Ratings
- Your TCI: na
- Joined: Thu Sep 17, 2015 9:36 pm
Re: Take into account the fact that not all users have seen the same films only
@uvlalid: welcome to the forum
@mpowell: I might be wrong, but I think you have misread uvlalid's proposal:
I don't think that's not how it is done currently. Criticker doesn't adapt the percentiles based on the target user for the comparison — which seems to be his suggestion. Criticker calculates the percentiles for user A and then compares user A's percentiles to user B's percentiles, user C's percentiles, user D's percentiles, etc.
I think uvlalid suggests that the percentiles should be calculated differently each time, taking into account the films in common. If you read his suggestion, that's not the same process as you use currently.
I have not given any thought on its merits — just saying that the proposed method is different from the currently implemented one.
@mpowell: I might be wrong, but I think you have misread uvlalid's proposal:
Using my suggested system to calculate the TCI of User 2 for User 1, only Films A and B will be used which for both users will have percentile scores of 100 and 0 respectively in that case.
I don't think that's not how it is done currently. Criticker doesn't adapt the percentiles based on the target user for the comparison — which seems to be his suggestion. Criticker calculates the percentiles for user A and then compares user A's percentiles to user B's percentiles, user C's percentiles, user D's percentiles, etc.
I think uvlalid suggests that the percentiles should be calculated differently each time, taking into account the films in common. If you read his suggestion, that's not the same process as you use currently.
I have not given any thought on its merits — just saying that the proposed method is different from the currently implemented one.
- livelove
- Posts: 759
- 67 Ratings
- Your TCI: na
- Joined: Thu Sep 17, 2015 9:36 pm
Re: Take into account the fact that not all users have seen the same films only
yes, but you use the films in common only when building the TCI.mpowell wrote:Hi! Yes, actually we do this already. Only the films in common are considered when building the TCI
He is talking about something else,
namely using the films in common when calculating the percentiles:
uvlalid wrote: instead of calculating the percentile scores of films based on all the films seen by a user, the percentiles should be based only on all the films both users have in common.
- livelove
- Posts: 759
- 67 Ratings
- Your TCI: na
- Joined: Thu Sep 17, 2015 9:36 pm
Re: Take into account the fact that not all users have seen the same films only
cross-reference:uvlalid wrote: instead of calculating the percentile scores of films based on all the films seen by a user, the percentiles should be based only on all the films both users have in common.
the same proposal has also been made in this thread:
Criticker ranking system has one enormous flaw
MTR wrote: you would need to calculate percentiles only among shared rated titles
- uvlalid
- Posts: 8
- 1339 Ratings
- Your TCI: na
- Joined: Tue May 15, 2018 6:21 pm
Re: Take into account the fact that not all users have seen the same films only
livelove wrote:@uvlalid: welcome to the forum
@mpowell: I might be wrong, but I think you have misread uvlalid's proposal:Using my suggested system to calculate the TCI of User 2 for User 1, only Films A and B will be used which for both users will have percentile scores of 100 and 0 respectively in that case.
I don't think that's not how it is done currently. Criticker doesn't adapt the percentiles based on the target user for the comparison — which seems to be his suggestion. Criticker calculates the percentiles for user A and then compares user A's percentiles to user B's percentiles, user C's percentiles, user D's percentiles, etc.
I think uvlalid suggests that the percentiles should be calculated differently each time, taking into account the films in common. If you read his suggestion, that's not the same process as you use currently.
Thanks for clarifying my position.
I have not given any thought on its merits — just saying that the proposed method is different from the currently implemented one.
I think that one benefit this would have is for users who, for whatever reason, don't manage to rate films that they would otherwise have rated with low scores (e.g., if they don't watch films that they think would be bad because they think it's a waste of time), therefore skewing their percentiles against films they think are good but have low percentile scores because they are ranked towards the bottom of their ratings. With my proposal, the percentile score of a film used in calculating TCI wouldn't be dependent on other films that the user has watched but that the other user (with which the TCI is being calculated) hasn't, i.e., films that are relevant to one user but not the other. Another way to look at it (from the perspective of the algorithm) is that a film would have a low score because it is worse than other films that both users have watched and not because it is worse than other films that only one user has watched, i.e., the algorithm would be less one-sided.