How Criticker fits into my personal movie recommendations...

Introduce yourself to the community or chat with other users about whatever is on your mind
Quicky
Posts: 451
0 Ratings
Your TCI: na
Joined: Mon Feb 25, 2008 2:53 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Quicky »

paulofilmo wrote:[...] Seriously, these results are stunning. They really look (and quite personally) like my kind of films. I'm going to create a 'Quicky' category in my IMDb 'My Movies'. [...]

I'm glad you seem to like it! Be sure to keep me informed on your experiences in the long run.

paulofilmo wrote:[...] Do you know why the first two films under 'Possibly not available for rent' have such high percentages? I'm consumed in the prospect that I could rate a film perfect. [...]

You shouldn't think of a movie having '100%' as perfect. My program normalizes the rankings such that the top ranked movie has '100%', the lowest ranked movie has '0%', and the movie that is ranked exactly in the middle has '50%'. A movie having '100%' only means it has the best chance at pleasing you. It's all about probabilities, really :).

paulofilmo wrote:[...] Would you be able to tell me what correlations/Final fit weights on the first page mean? I think I'll be able to decipher the dots, dashes and arrows by comparing the films. [...]

I guess I should make a fourth page with some explanations, eh? :)

Ok, let me explain correlations with two examples.
1) Let's say you order 10 cars by their size, along with their price tag. I guess you'll agree with me that generally the bigger the car, the more expensive it will be. But I'm sure you'll also agree that that is not always the case, since the price depends on more than just size, for example: luxury accessories, brand, etc. What this means is that the correlation between a car's size and its price tag will be somewhere between 0 and 1. If it were 1, the correlation would be perfect and a bigger car would always be more expensive than a smaller one. If it were 0, you would have no way of telling whether a car would be more expensive based on its size.

2) Let's say you order 10 microchips by their size, along with their price tag. Unlike the first example, here the chips will usually be more expensive when they're smaller. So the correlation between a chip's size and its price is negative, somewhere between -1 and 0.

Here you can find a picture that shows graphs of variables with different correlations. I would guess that my two examples will be best illustrated by the lower two graphs. The left one being the cars, the right one being the chips.

So in my program I test how well the correlation is between your ratings and certain characteristics of a movie that can be quantified. If your PSI correlation would be 1, this would mean that if movie A has a higher PSI than movie B, you will always rate movie A higher than movie B. And if your Age correlation would be -1, this would mean that if movie A is more recent than movie B, you will always rate movie B higher than movie A.

These correlations only show how your ratings are influenced by each parameter on its own. It's very hard to determine how your ratings will be influenced when you combine all of these parameters, because there are also correlations between the parameters. For example, there will usually be a pretty strong correlation between your PSI's and your favourite genres, because your PSI's will already favour movies with those genres that you like.

My program tries to find the combination of weights for the parameters that gives the best fit to your ratings. So the percentages that you see in the right column show how heave each parameter counted towards a movie's ranking in the system.

I hope that made it somewhat clear?

Quicky
Posts: 451
0 Ratings
Your TCI: na
Joined: Mon Feb 25, 2008 2:53 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Quicky »

Magb wrote:Wow, I'm impressed. The effort you're putting into this is way beyond the call of duty. It would be nice if Criticker was a little more accomodating for your purposes, for instance by showing users' PSI's for movies they've already rated. If you would check out my account any time you feel like it then it'd be greatly appreciated. (PM sent.)

Ok, Magb, I'm finished with yours as well: http://thequicky.net/files/magb-qmdb-re ... ations.pdf

As you will see in the correlations column, you're clearly not into blockbusters, which makes it a pain in the ass for me to add all those obscure movies, including a few dozen disney shorts from the early 20th century :D. Ah well... it's still fun to do ;).

paulofilmo
Posts: 2586
5 Ratings
Your TCI: na
Joined: Mon Apr 07, 2008 3:40 pm

Re: How Criticker fits into my personal movie recommendations...

Post by paulofilmo »

I have trouble putting the final fit weights into relative perspective. I really don't have a head for numbers. But thank you for your explanation, I'll read through the thread a few more times.

Quicky
Posts: 451
0 Ratings
Your TCI: na
Joined: Mon Feb 25, 2008 2:53 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Quicky »

paulofilmo wrote:I have trouble putting the final fit weights into relative perspective. I really don't have a head for numbers. But thank you for your explanation, I'll read through the thread a few more times.

Let me say one more thing. The first column of correlations isn't used in any of the calculations that yield your list of movies. I only calculate them to illustrate how each parameter works on its own compared to your rating.

The final score for each movie is a combination of the scores in each parameter. However, each of these parameters is weighted so that the end result is closest to your ratings. PSI is usually the most important parameter and thus has the largest weight in the summation, while Age and Length only have minor weights. The reason that high weights don't always correspond to high correlations is because of the correlations between those parameters that can't be quantified easily. My program just tries to find the best set of weights by brute force (testing in total 5 times 20,000,000 sets of weights).

The important thing to realize is that the program tries to use the available parameters in such a way that it comes up with a ranking that approaches your ratings as close as possible. That's it really :geek: .

Magb
Posts: 47
0 Ratings
Your TCI: na
Joined: Tue Mar 21, 2006 11:40 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Magb »

Quicky wrote:
Magb wrote:Wow, I'm impressed. The effort you're putting into this is way beyond the call of duty. It would be nice if Criticker was a little more accomodating for your purposes, for instance by showing users' PSI's for movies they've already rated. If you would check out my account any time you feel like it then it'd be greatly appreciated. (PM sent.)

Ok, Magb, I'm finished with yours as well: http://thequicky.net/files/magb-qmdb-re ... ations.pdf

As you will see in the correlations column, you're clearly not into blockbusters, which makes it a pain in the ass for me to add all those obscure movies, including a few dozen disney shorts from the early 20th century :D. Ah well... it's still fun to do ;).

Thanks! The results are more or less what I expected, although I suspect that some of my correlations -- especially "Age" -- are somewhat skewed because, as you mentioned, I've ranked a lot of old Disney short films and some other relatively inconsequential stuff that clearly doesn't have a major bearing on my general taste in movies.

While we're discussing these things I'd like to talk a bit about something that I've unsuccessfully tried to put into words before, but which I now think I understand better. I think most mathematical representations of subjective taste in film (or anything, for that matter) are going to be somewhat hampered by the fact that there's no practical way to determine how much the score of any given film should count in the total analysis. As I see it there are (at least) two different scores that you can give for a movie you've seen. One is the obvious one: how much do you like the movie, or how good do you think the movie is? The other is more subtle: how much do you care about how much you like the movie? For instance, I've given the Ken Burns documentary "The Civil War" a score of 100 because I think it's virtually flawless, but I'm nowhere near as enthusiastic about that film as I am about a film like "2001: A Space Odyssey" or "Casablanca". On the other end of the scale I've given "Freddy Got Fingered" a 0 because I think it's completely terrible in every way, but I wouldn't really hold it against someone if they loved it the way I would with a film like "The Day After Tomorrow" or "Die Another Day". Finally, in the middle of my ratings list I have both films like "Vacation" that I'm just completely indifferent about and films like "Titanic", which I strongly believe are neither good nor bad but aggressively average and mediocre.

The problem here is that a score of 100 for "The Civil War" and a score of 100 for "Casablanca" are counted equally in the development of my taste profile, when in fact I feel quite differently about the two in what appears to be a relevant way. This is true for all the ratings system I've encountered, and I can't think of a good way to avoid it. I guess the bottom line is that the mathematical measurement of subjective taste is an inexact science because subjective taste is so complex.

On another note, I find it funny that we all see to hate helicopters. Helicopters are cool with me so I don't know what that's all about.

Quicky
Posts: 451
0 Ratings
Your TCI: na
Joined: Mon Feb 25, 2008 2:53 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Quicky »

Magb wrote:Thanks! The results are more or less what I expected, although I suspect that some of my correlations -- especially "Age" -- are somewhat skewed because, as you mentioned, I've ranked a lot of old Disney short films and some other relatively inconsequential stuff that clearly doesn't have a major bearing on my general taste in movies.

The results could indeed be a little skewed in this regard, however at only 3.2%, the inherent skewing in the age parameter shouldn't have had too much of an influence.

Magb wrote:[...] While we're discussing these things I'd like to talk a bit about something that I've unsuccessfully tried to put into words before, but which I now think I understand better. [...] As I see it there are (at least) two different scores that you can give for a movie you've seen. One is the obvious one: how much do you like the movie, or how good do you think the movie is? The other is more subtle: how much do you care about how much you like the movie? [...]

Putting abstract things like movies and especially taste in movies into numbers is a very inexact science indeed and it will never be perfect. With my algorithms, I'm trying to make my recommendations as reliable and useful as possible, but it'll never be exact :).

Magb wrote:[...] The problem here is that a score of 100 for "The Civil War" and a score of 100 for "Casablanca" are counted equally in the development of my taste profile, when in fact I feel quite differently about the two in what appears to be a relevant way. This is true for all the ratings system I've encountered, and I can't think of a good way to avoid it. [...]

I think you make an excellent and very important point here, and I fully agree with you.

I remember this has come up in the past on the forum, when people said that they wanted to make their top 10 tier more important in calculating TCI's. As you very well showed, it's not just your top 10 tier that should be more important. You can have strong opinions about movies from every tier. I think it could actually be very useful and informative for Criticker to find a way to deal with this issue, and maybe this deserves a separate thread to brainstorm about this? :)

I think you could actually set up a separate rating for how important you think a movie is in your ratings. Of course this shouldn't be a continous rating from 0 to 100, but maybe just a simple flag (yes or no), or a few stars (0 stars, 1 star, 2 stars,...). In calculating your TCI's this information could then be used by doubling or tripling or ... the weight of the tier difference for this particular movie. You worded this really well, Magb, maybe you can start a new thread with your thoughts on this? :)

Magb wrote:On another note, I find it funny that we all see to hate helicopters. Helicopters are cool with me so I don't know what that's all about.

Indeed, I've also found this quite odd. Honestly, I have no idea where it comes from :P.

paulofilmo
Posts: 2586
5 Ratings
Your TCI: na
Joined: Mon Apr 07, 2008 3:40 pm

Re: How Criticker fits into my personal movie recommendations...

Post by paulofilmo »

I think docus have a handicap in not being able to create the ambiance of a regular film. When the profound personal connection is made with an abstract film or with a quality drama, something didactic seems to lose its relative value.

Quicky wrote:
paulofilmo wrote:Yea, you got it! Can't wait to see what your system comes up with. [...]

I just finished with yours :).

You can download your recommendations list with some statistics on the first page at http://thequicky.net/files/paulofilmo-q ... ations.pdf

What I'm observing for you is that your PSI's aren't doing as good a job as with the others. I don't really understand why though. Maybe your taste is too obscure? :-S Either way, I'm too tired right now to take a deeper look.

Magb, I'll do yours tomorrow :).


My PSIs have a poor correlation, but a stronger percentage weight than most people. Is this because the other factors are also weak?

Quicky wrote:I'm glad you seem to like it! Be sure to keep me informed on your experiences in the long run.


I'm eying up some relevant DVDs, and will let you know how I get on.
Last edited by paulofilmo on Wed Apr 23, 2008 9:17 pm, edited 1 time in total.

MmzHrrdb
Your TCI: na

Re: How Criticker fits into my personal movie recommendations...

Post by MmzHrrdb »

Ready to do mine yet? :D

If it doesn't involve much work for you, I'd like to know what my 10 top/bottom keywords are.

Quicky
Posts: 451
0 Ratings
Your TCI: na
Joined: Mon Feb 25, 2008 2:53 pm

Re: How Criticker fits into my personal movie recommendations...

Post by Quicky »

(I'm sorry, I posted this under my girl friend's account at first.)

FitFortDanga wrote:Ready to do mine yet? :D
If it doesn't involve much work for you, I'd like to know what my 10 top/bottom keywords are.

I can't do only the keywords, but I can make a recommendations list for you, with the restriction that it'll only be based on those movies that I have in my database. It'd probably take me about 20 or so hours to add yours to my database. At €10 per hour that'd be €200 or $320, right? :P... Nah just kidding, but I'm sure you get my point ;).
Anyway, so if you want to give that a try, just send me your password in a private message and I'll give it a whirl with the data that I do have.

td888
Posts: 836
0 Ratings
Your TCI: na
Joined: Thu Jul 12, 2007 10:44 am

Re: How Criticker fits into my personal movie recommendations...

Post by td888 »

Quicky wrote:Anyway, so if you want to give that a try, just send me your password in a private message and I'll give it a whirl with the data that I do have.


Can you give mine a swirl again? I send you my pw via PM... :)

Post Reply