A new reddit voting system?

With the discussions about the quality1loss2 of reddit submissions and comments, I thought about a possible way of changing the voting system.
As a user, you currently have one upvote/downvote. This is obviously problematic, because not everybody views the same thing as good. You would have to somehow categorize what you like in different areas, so multidimensional upvotes? Sounds like a bad idea.
But you don't have to present this to the user. So, keep the current upvote/downvote arrows, but instead of directly adding them up, give posts an n-dimensional vector as rating. The user will recieve one as well, bound to his account and describing what he likes. The system can then compare the two vectors, a high similarity should indicate content the user wants to see, this will give it a high rating for this user. This could be done by projecting the vector of the post onto that of the user and then determine its length. The higher the value, the better.
How are these vectors generated? When a user votes on a post, his voting vector gets imprinted into the post vector, in the simplest case by just adding them up. At the same time, the post vector gets imprinted onto the user vector, for example by norming the length of the post vector to one and then taking the weighted average. To help set things up, a small random offset should be added to he user vector before adding it into an upvote.
This approach has of course some implementation problems, namely the by far increased computational costs (I don't know how much they take for the current reddit system).

I hope this is understandable, I am pretty much just an interested layman in machinelearning.