[what]

I read somewhere about a recommender system for movies (I think) and what they did is force a user to rate 5 random movies as part of the registration process. The movies weren't entirely random, but ones they thought were significant in identifying a user's tastes.

In your example of browsing patterns, maybe you could ask new users if they do or do not like to read certain types of articles. ie: are you interested in technology, sports, entertainment, random pictures of cats etc and seed their profiles based on their expressed level of interest for those things (maybe including dislikes from people who claimed to have similar interests).

But I would think that dislikes are not so important in the beginning. Although I don't know how your algorithm works, if you have a rough idea of what a person likes, shouldn't you be able to recommend things that they might like just based on that? When you end up recommending something that they don't like, you'll get some dislike data and can start factoring that in.