Blog, 2015

I added two new features to the recommendations :)

First, you can filter your recommendations based on year. Like in the image below, where I'm apperently really in the mood for an action movie from the turn of the millenium.

Second, you can now ignore recommendations. If you've done you own research and decided Pearl Harbor isn't gonna be your cup of tea, just click on the X to the left to ignore it.

By comparing movie watchers with each other Veboli can calculate really accurate recommendations. But to be able to compare movie watchers with each other users have to spend some time telling Veboli everything they like and dislike. I thought that that was just how it was, a time investment to get really good recommendations.

When some friends asked me if I could speed up the process I did some thinking. Annnddd (do you feel it coming) of course I could! Veboli already has a smart algorithm for comparing movies to each other, so I just had to incorporate this algorithm into the recommendation algorithm.

Now you just have to give me one of your favorite movies and you'll get a bunch of recommendations. And you can get more and more accurate recommendations any time by rating more movies.

Signing up is also super easy!

The other day, Lenneke van Rossum told me I should make the home page sexy. Good advice, but I thought that it would be way too much work. Still, I went home and played around with the idea.

After two days, it was almost done! Just some small adjustments and the home page was sexy:

Only the necessary information, and with nice images from well known movies. I'm really happy with it :)

So I finished transferring everyone's song ratings to the new Musicbrainz database. Great news, but one little caveat. Only 20% of the songs had equivalents between the old Freebase data and the new Musicbrainz data!

With Freebase I used a nice and complicated system to figure out what was a unique song (all the different live versions of a song aren't really different songs, but a cover by a different band definitely is). With Musicbrainz I simplified it by looking at works instead of recordings, but works is a rather incomplete dataset compared to recordings, hence only some of the songs have equivalents.

To get the rest of the song ratings transferred I'll make use of the recordings as well. Time to look at the old complicated system again :)