Geometry, Data and Neighbors Predict People’s Favorite Movies
quantamagazine.org
Geometry, Data and Neighbors Predict People’s Favorite Movies
1–5 of 5 posts
Re: Geometry, Data and Neighbors Predict People’s Favorite Movies
#2The article holds up Netflix as a shining example of movie recommendation engines but Netflix has only 3000 to 5000 distinct movie titles (depending on your country). You cannot get interesting or surprising movie recommendations from such a small search space. Imagine the extreme case where Netflix had 1 single movie, say "Twilight". They'd always recommend "Twilight" even if Donald Knuth wrote their recommendation engine.
You need something like IMDb's catalog of 500,000 movie titles to do something interesting.
Re: Geometry, Data and Neighbors Predict People’s Favorite Movies
#3The article holds up Netflix as a shining example of movie recommendation engines but Netflix has only 3000 to 5000 distinct movie titles (depending on your country). You cannot get interesting or surprising movie recommendations from such a small search space. Imagine the extreme case where Netflix had 1 single movie, say "Twilight". They'd always recommend "Twilight" even if Donald Knuth wrote their recommendation…
It also stated that for a larger set is complex and not that easy that's why the article holds up Netflix as a shining example.
Re: Geometry, Data and Neighbors Predict People’s Favorite Movies
#4The article holds up Netflix as a shining example of movie recommendation engines but Netflix has only 3000 to 5000 distinct movie titles (depending on your country). You cannot get interesting or surprising movie recommendations from such a small search space. Imagine the extreme case where Netflix had 1 single movie, say "Twilight". They'd always recommend "Twilight" even if Donald Knuth wrote their recommendation…
MovieLens [0], by far, my favorite movie recommendation engine, says that the dataset it was based on [1] contained "2,811,983 ratings entered by 72,916 [users] for 1628 different movies". By comparison, its current biggest published dataset [2] contains "20 million ratings and 465,000 tag applications applied to 27,000 movies by 138,000 users. Includes tag genome data with 12 million relevance scores across 1,100 tags".
On a side note, if you like discovering movies, I can't recommend MovieLens enough. Its accuracy and ability to recommend obscure movies still blows me away. On top of all that it's free, as in no subscriptions or ads, and it's developed and maintained by "GroupLens, a research lab in the Department of Computer Science and Engineering at the University of Minnesota".
[0] https://movielens.org/ [1] https://grouplens.org/datasets/eachmovie/ [2] https://grouplens.org/datasets/movielens/
Re: Geometry, Data and Neighbors Predict People’s Favorite Movies
#5How are the data points assigned to their locations in the geometric space? I didn't notice any mention of that. What makes something "more like" a thriller and "less like" a comedy, for example? We can all tell the intuitive difference, but what markers serve best to precisely assign a mathematical distance?
I know very little about this problem space, so these are actual questions, not at all rhetorical.