In another tread, td888 wrote:[...] Hi Quicky, how are my 'personal recommendations' going? Lot of work...?
I just finished
.And moreover, once again I added a new parameter to my calculations. Actually I added one and removed one (Awards) because at this very moment, my code doesn't allow more than 7 parameters. I hope to find a way around this in the future. The parameter I added is keywords. I now make a list of keywords that are mentioned on the IMDB-keywords pages of the movies you have rated. Using this list, I try to predict whether you'll like other movies based on its keywords. Some keywords will be new, some will not.
[WARNING: ABSTRACT MATH CONTENTS]
Now... I can choose how many keywords to use for this. The more keywords, the better the system will predict which movies you liked. However, it would be very unwise to use too many keywords, because then you'd be approaching a scenario where you say "Is the name of a movie 'There Will Be Blood"? If yes, then I'll like it, if not, then I won't. Obviously this would mean that my program correctly predicts that you like "There Will Be Blood". But I'm sure you'd agree that nevertheless, this program doesn't necessarily give informative predictions for movies that are not "There Will Be Blood". It's important to realize that to get accurate predictions, you need to find a balance between criteria that are too general and too specific. Simply decide which movie you're going to watch based on whether it's a comedy or a not is an obvious example of being too general. Using the title of a movie as a criterium would be too specific because there will only be one movie with that title. In the same way, using too many keywords would mean there will only be very few movies that have the keywords you liked, i.e. the ones from the movies you liked and have already seen.
I'm sure I've lost most of you by now....
[/WARNING]
td888, you can find your recommendations list at http://thequicky.net/files/td888-qmdb-r ... ations.pdf
Your ratings correlated to each of the 7 parameters as follows:
(1 means perfect correlation, 0 means no correlation at all, -1 means perfect but opposite correlation)
- PSI: 0.42 (PSI is helpful, but it's not a great fit)
- Popcorn: -0.15 (like everyone else here apparently, you prefer smarts over popcorn)
- IMDB: 0.42 (IMDB helps with your recommendations)
- Length: 0.27 (you prefer longer movies)
- Age: -0.13 (you have a very slight preference towards older movies)
- Genre: 0.28 (genre is pretty important to you. you love documentary, and history, but you hate comedy and family)
- Keywords: 0.14 (the keywords parameter does help)
In order to give you the best fit to your personal preference the following weights were used to come up with your recommendations:
- IMDB: 29.1%
- PSI: 26.3%
- Genre: 18.3%
- Keywords: 11.5%
- Popcorn: 8.5%
- Length: 4.1%
- Age: -2.3%
Oh and, to give you some insight into the keywords, here are the 20 keywords you feel strongest about:
- Top 10 (keywords you like): Tragedy, Neo-Noir, Violent Movie, Disturbing, Psychopath, Famous Score, Brutality, Critically Acclaimed, Vengeance, Bloody Violence
- Bottom 10 (keywords you don't like): Sequel, Character Name in Title, Fall from Height, Female Nudity, Helicopter, Police, Kids & Family, Dog, Father-Son Relationship, Twist in the End
Apparently you like brutal and bloody tragedies about psychopaths wanting vengeance that are critically acclaimed and have famous scores
. And you seem to hate sequels that have nude women falling out of helicopters being chased by police dogs and having a twist in the end... 
Now, back to seriousness... I noticed that none of the parameters really give a good fit to your ratings, causing the final fit also not to be that great. I strongly suspect the reason behind this is that you have a rather unvaried system for rating your movies. You rated roughly 25% of your movies as '75' and roughly 20% of your movies as '70'. This means that my system doesn't have a good way to approach your ratings. I would suggest that you differentiate more in your ratings, using the full scale from 0 to 100 and trying to repeat ratings as little as possible. With almost 1100 movies that could be a pretty big task, I realize, but I think you will get better results both with Criticker's PSI as well as my recommendations system if you try to do this.
My own top 20 keywords turned out to be the following:
- Top 10: No Opening Credits, Murder, Shot in the Head, Blood Splatter, Epic, Based on Novel, Female Nudity, Violence, Tragedy, Flashback Sequence
- Bottom 10: Humor, Kids & Family, Character Name in Title, Dog, Remake, New York, Person on Fire, Brother-Sister Relationship, New York City, Mother-Son Relationship
Apparently I like epic movies based on novels that have nude women being shot in the head during a flashback sequence. And apparently I hate comedy remakes about kids being on fire in New York City and having a relationship with their brothers and sisters. Quite insightful, eh?
