Ratings: Algorithm to predict how I'm going to score a film would be great.

  • 6
  • Idea
  • Updated 7 years ago
  • (Edited)
I think there could be an algorithm created that would predict what a person would rate a film if they haven't rated it yet. This would help them find titles that they would likely enjoy. Factors to be taken into account would be how the person has rated genre films in the past, the imdb score and particularly how users with similar taste have rated the film, and how the person has rated films with the cast and crew in the past. For example I am lookoing up a Martin Scorcese film. On average I rate them 1 point lower than the average user (don't know if I actually do, just making up for the example). The film has an IMDb score of 8, therefore all other things being equal it should predict that I will score the film 7. Or I am looking up a horror comedy with a score of 4.5 and on average I rate them 1.5 points higher than their IMDb scores. Therefore all other things being equal it should predict a score of 6 for me.
Photo of Galen Mountfort

Galen Mountfort

  • 14 Posts
  • 8 Reply Likes

Posted 8 years ago

  • 6
Photo of Emperor

Emperor, Champion

  • 6418 Posts
  • 3026 Reply Likes
You could take into account things like genre and/or director or leading actors.

You could then sort the recommendations by this. If I liked John Carpenter's horror movies and tended to rate them 2 stars over the average then they might suggest others of his films.

At the moment it seems to work that if you liked X then you might like Y, which seems a little crude and isn't overly accurate either.
Photo of Dan Dassow

Dan Dassow, Champion

  • 17172 Posts
  • 19619 Reply Likes
A few years ago, IMDb linked to a site that used collaborative filtering [http://en.wikipedia.org/wiki/Collabor...]. Unfortunately, IMDb no longer links to that site; that site may also no longer exist.

One major difficulty with using IMDb rating data to predict your rating is that IMDb has no way of verifying that a user has seen a film. IMDb can and does filter out obvious ratings fraud, but it is probably not feasible to ever completely validate IMDb's data set.

Finding other users with a similar rating history to you would be very helpful, since IMDb allows users to compare their ratings with public ratings.

In contrast, Amazon.com and Netflix use purchase and viewing information for their recommendation algorithms, and generally have better results.

You may find the following articles interesting.
* Netflix Prize [http://en.wikipedia.org/wiki/Netflix_...]
* Why Netflix Never Implemented The Algorithm That Won The Netflix $1 Million Challenge [http://www.techdirt.com/blog/innovati...]
* Algorithms: The Ever-Growing, All-Knowing Way Of The Future [http://www.npr.org/blogs/alltechconsi...]