QMDB Recommendations - Part 2

[KGB]

I don't think that everybody follows my system... I just don't think the great films can be appreciated without seeing some really awful or simply mediocre film for comparison. Also, it mantains my rankings list from my tiers going way off my quips.

[Quicky]

I don't think that everybody follows my system... I just don't think the great films can be appreciated without seeing some really awful or simply mediocre film for comparison. Also, it mantains my rankings list from my tiers going way off my quips.

Ah ok, I understand. Though I think it'd be a bit useless to add awful movies to the mix. I think in that case it is best left to the user, i.e. you, to decide what to watch in between the movies from my list . There are so many awful movies out there, I don't have a clue how I'd best pick one .

But I'm glad you like the new varied recommendations!

[KGB]

That was my point, the varied recommendations still serves as an excellent guide to choose a great film to watch.

[paulofilmo]

Thank you, Quicky. I've hiked up my 'Films in Common Min' percentage, and I think it's giving me better results, so this is likely the most accurate list so far. Although, I think some of my PSIs may have changed quite recently as they differ quite a bit. i.e. I saw "All About Eve" yesterday, which has a PSI of 93. Regardless, your QMDB prediction was very accurate.

I should be able to see more of those films soon.

I have should Veoh links to a few films on other peoples' lists -- (Godfather, City of God, Diving Bell, Rashomon, The Battle of Algiers, Holy Mountain, Tokyo Story, Grave of The Fireflies, Spirited Away, In The Mood For Love, This Is England, Princess Mononoke, Wild Strawberries, American History X, Casablanca, Lawrence of Arabia, Double Indemnity, It's a Wonderful Life, 3-iron, Reservoir Dogs, MP and the Holy Grail, Schindler's List, 12 Angry Men, Band of Brothers [parts 1-8, Jap subs], Saving Private Ryan [Jap subs], Infernal Affairs [no subs], American Splendor)

Send me a pm if you can't find one of those yourself. Don't get your hopes up as they may have been removed. Let me know if you want me to keep an eye out for anything specific.

Cheers

paul

[Quicky]

Here's another update. Instead of the Accuracy Percentage on the first page of the .pdf files, I'm now showing how well my database performs by calculating a TCI similar to how they're calculated here on Criticker. The only difference is that I put more weight on the better movies. The reason for this is that I don't care so much how accurate my database is in the low end. My lists don't contain the low end anyway. I want my lists to be accurate on the high end of the scale. This is actually something I believe would make Criticker even more useful as well if they'd implement it like that.

For myself I got the following results:

My best user TCI as mentioned on my 'Your TCI' page: 1.67
Performance of Criticker's PSI's: 1.63
Performance of my database: 1.59

Read the following if you're bored yet awake: ;)
So for me my own system using Criticker's data outperforms Criticker, though not by that great a margin. I was actually a bit disappointed by that . But anyway... I guess it shows how good Criticker actually is , but I also think it shows that Criticker has a lot more potential than it is showing today. Some of you might be a bit surprised that the PSI's are actually not that much better than your best user TCI. This surprised myself a bit as well. The advantage of your PSI's is of course that you have PSI's for thousands of movies, whereas those users with good TCI's usually have rated only a couple of hundred movies.

I've done these calculations for all of you (the ones for which I still have passwords), so if you want it, let me know and I'll upload it again. For some people my own system outperforms Criticker by a larger margin (up to 0.25 or something like that), for one user Criticker actually outperforms my own system. So the performance depends heavily on the user.

My system in itself performs best for KGB with a TCI of 0.96 (best user TCI of 0.95) and worst for paulofilmo with a TCI of 1.82 (best user TCI of 1.81). As you can see for them their best TCI actually outperforms the whole of Criticker and my own system. This shows that their lowest TCI's are so good that they adding more users actually brings down the accuracy, though of course with the advantage of having PSI's for a bigger set of movies. In general I think this shows how valuable the reviews of your very lowest TCI's are.

Compared to Criticker my system performs best for jeff_h with a TCI of 1.61 where the Criticker TCI is 1.80 (best user TCI of 1.40), and it performs worst for Llamadeus with a TCI of 1.01 where the Criticker TCI outperforms my own system with 0.96 (best user TCI of 1.02).

[Quicky]

MAJOR UPDATE!

-- DIRECTORS / ACTORS --
I am now not only taking keywords into account, but also directors, screenwriters and actors/actresses. This new parameter is combined with the keywords parameter and together I now call it the tags parameter. For KGB (see below) this resulted in an increased accuracy of TCI 0.93 (it was 0.96 before). In the pdf's the keywords column has now been renamed to Tags and that the icons in this column now show whether you like the keywords and the people involved with this movie. Aside from the sometimes silly top 10 and bottom 10 keywords, I'm now also showing your top 15 and bottom 15 directors/actors. For the moment I'm not yet distinguishing between directors and actors, so it's possible that you'll only see actors in your top 15. It'll take quite a bit of extra coding to distinguish them, and I've done enough coding for now .

-- RATINGS THROUGH THE AGES --
Furthermore, I've removed the page that had the genres split up since I personally don't see the use of it anymore now that I have the varied recommendations list, and instead I now added a page with extra information. This includes a ratings evolution from 1920 till 2008. So here you can find out if you really love the old movies more than the new ones and you can also gasp at the lack of old movies you've seen compared to the 90's and 00's.

-- SOME EXTRA STATISTICS --
Another new thing is that I've also added some statistics that I was already using to determine your recommendations, but were not visible till now. These simple statistics show you how much you watch each of the 5 main genres, and what the usual popcorn rating is for the movies that you watch, as well as the median year of release. So if you usually watch recent action flicks, this will be reflected in these statistics. However, it is not because you watch a lot of action movies, that you also rate these movies high. In fact I've noticed that several people watch a lot of action and humor but rate drama and arthouse the highest. This means that in my varied recommendations I try find the middle ground between what you usually prefer watching and what you rate the highest.

-- ANIMATION GENRE --
I made one small improvement in the main genre categorization. From now on, I'm not counting the Animation genre anymore when determining the main genre category of a movie. This means that animated movies will now not almost automatically be placed into the 'Humor' genre category, but will now end up wherever their other genres lead them. In most cases this is still the humor category, but for example, Grave of the Fireflies now ends up in arthouse instead of humor.

The first person I tried the new version on was KGB: http://thequicky.net/qmdb/kgb-qmdb-reco ... 080624.pdf

As you can see this confirms that two of KGB's favourite directors are Stanley Kubrick and Werner Herzog. It also shows that he prefers movies from ca. 1930 to ca. 1975, and that the couple of years around 2000 were a definite low-point in good movies for KGB. And despite the fact that he watches a lot of action and humor, he rates those the lowest. Of course you do have to keep in mind that we are all probably pickier when choosing a movie to watch from before 1970 than we are when we're watching a recent one on TV...

[td888]

Can you generate one for me? I just send you my pass.

[KGB]

Oooh, more QMDB list for me! Thanks!

But I don't understand one thing, though - what is the 'Watching Habits' thing all about? I didn't entirely understand the explanation...

[PeaceAnarchy]

Oooh, more QMDB list for me! Thanks!

But I don't understand one thing, though - what is the 'Watching Habits' thing all about? I didn't entirely understand the explanation...

It tells you how often you watch each type of movie, regardless of how much you like them. So from your example, even though only 26% of the movies you like are action and comedy, they're 54% of what you watch.

This is really cool, and I hope you can do one for me as well. I especially like your idea of calculating a TCI for both your recommendations and criticker's.

[Quicky]

Can you generate one for me? I just send you my pass.

This is really cool, and I hope you can do one for me as well.

Working on it

It tells you how often you watch each type of movie, regardless of how much you like them. So from your example, even though only 26% of the movies you like are action and comedy, they're 54% of what you watch.

Exactly. I've changed the explanation a bit in the .pdf.

I especially like your idea of calculating a TCI for both your recommendations and criticker's.

I'm still not sure whether I'm happy with the fact that in some cases my own TCI is higher than Criticker's TCI, but well... It should give you an idea at least of how accurate your recommendations really are..