QMDB Recommendations - Part 2

[paulofilmo]

Ah right. I think I get it now. I first read that, looking at it blankly; non-the-wiser.

So for you (for example), 1997 is the median where the films have a good ratio between the age of a film you choose to watch and also the kind of films you think highly of. That's understandably difficult to put into words, but I think I get the gist of it.

[Quicky]

Exactly

Note: I will be on vacation from Tuesday 8th till Thursday 24th of July, so I will not be here to make any lists for people. If you want (an updated) one, just let me know and I'll get to it when I get back!

[Quicky]

A first recommendations list for LordofDance: http://thequicky.net/qmdb/lordofdance-q ... 080918.pdf

[krmr]

I'm a bit of a lurker on the forums, but I'd love to get the QMDB treatment. I was actually attempting to do something similar with my own database of ratings, but yours is much bigger in scope and better because of it.

[Quicky]

There have been some updates to my QMDB algorithm over the summer.

• First of all, there was a bug which switched the parameter weights of two parameters which could cause inaccurate results for those people who had very different weights for these parameters. This has now been fixed, so my QMDB is performing better now than it was before. I found the error when I was brainstorming about possible reasons for my QMDB to perform worse than Criticker which is theoretically impossible since my QMDB algorithm optimizes the weights such that it reaches the best fit, which if Criticker would be perfect would converge on a weight of 100% for the PSI's. Apparently this flaw was caused by the swap of these two parameter weights.
• Secondly I have now split up the names parameter properly into directors/writers and actors. This means that on the first page of your QMDB list, you now have top/bottom lists for director/writers, actors and keywords separately.
• Lastly, I decided to not exclude movies from the varied top recommendations list that are currently a new release according to Criticker.

I have made new lists for paulofilmo and kramertron which incorporate these updates:

• paulofilmo: http://thequicky.net/qmdb/paulofilmo-qm ... 080923.pdf
• kramertron: http://thequicky.net/qmdb/kramertron-qm ... 080924.pdf
• Quicky: http://thequicky.net/qmdb/tijl-qmdb-rec ... 080923.pdf

If you want an updated list (also correcting the bug listed above), or you have never asked for a QMDB list before but are interested in getting one, please let me know either here in the thread or by PM.

As always, I can only improve these lists if I get feedback from you on what you like and what you don't like and why, and why you use or don't use the list in practice .

[Quicky]

Current performances*:

• Quicky: Lowest 3 User TCI's: 1.63, 1.69 and 1.70, Criticker TCI: 1.52, QMDB TCI: 1.32
• paulofilmo: Lowest 3 User TCI's: 1.90, 1.90 and 1.92, Criticker TCI: 1.99, QMDB TCI: 1.77
• kramertron: Lowest 3 User TCI's: 1.21, 1.59 and 1.72, Criticker TCI: 1.69, QMDB TCI: 1.39

As you can see, my QMDB system improves Criticker's recommendations by lowering the TCI another 0.2 to 0.3 points for these three people, meaning it improves the accuracy of the recommendations by roughly 15%.

*Note: The Lowest 3 User TCI's are the Taste Compatibility Indexes (lower is better!) for the top 3 users here on Criticker that are closest to your taste. The Criticker TCI is a TCI-value** for how closely Criticker's PSI's resemble your actual taste in movies, i.e. how good Criticker is at guessing whether you'll like a movie or not. The QMDB TCI is a TCI-value for how closely my QMDB system resembles your taste in movies.

**Note: for the Criticker and QMDB TCI's, I use a fitting algorithm that is skewed to the best movies, as opposed to the way Criticker fits tastes uniformly over the spectrum of movies from good to bad. The reason I do this is simply because I believe it is a more helpful way to compare movie tastes.

[paulofilmo]

Great job on getting my best TCI, Quicky. And thank you for another update; intriguing stuff.

I actually have 12 of my varied top recommendations and 5 of my 'more recommendations' immediately available to me, so I'll put some focus on those.

LOL at Ethan Hawke being in my top 10 actors. I don't like him, but he picks some fine scripts.

My key words are spot on, excitingly so. I'll introduce a 'Naturalism' keyword if there isn't already one.

[Quicky]

Spunkie asked me to do a list for him, so here goes: click!

[Spunkie]

Yay, cheers ! After an examination it's easy to see your system works quite good.....except :KEYWORDS Your Bottom 10 helicopter. I have no prejudice against choppers

This is ironic
EXAMPLE TIERS
T5 Good Superbad (2007)

[paulofilmo]

Quicky, I've seen about four of the films from your recs pdf. Is there any specific input you're after, or are you after ideas?