I can't comment much on this problem. It seems very interesting and I'd love to dive into the implementation. One place I think you might want to look for a similar problem is Netflix recommendations. Netflix specifically takes your ratings into account to tailor their recommendations for you. Sometimes I'm impressed with what they can come up with. Other times it seems way off base, but I'm constantly exposed to interesting and new things through their recommendations. I'm also very impressed with how quickly they identify categories of things I'm interested in that are surprisingly specific.
The Netflix recommendation system is definitely not solved yet, but they have had competitions to improve it and you could probably get some good ideas about how to use a user's history to predict their future preferences.
There was actually a timely article a little while ago on Wired (http://www.wired.com/underwire/2013/08/qq_netflix-algorithm/) that talked about the Netflix recommendation engine and how it has had to evolve since the days of the Netflix Prize. Interestingly (to me) was that they have actually had to move away from ratings comparisons with the shift to a primarily "on demand" style of viewership. As there is so little personal investment in a choice when viewing streamed content (you can just switch 5 seconds in) people are far less likely to reliably rate material. Instead, they've had to develop algorithms which rely primarily on recent viewing habits (clicks, content seen to completion, left early, time of the day) for their recommendations. Which then sounds very similar to the goal of recommending viewership topics based on prior reading, rating, and participation in threads. However, as was noted above, there's the issue of "genetic diversity" to your viewing pool, so you'd probably want to do something like Pandora and cross-breed your interests with those of similar, but non-identical viewers to suggest topics that might interest you. Giving a profile of how they are balancing you that you could also edit would help further.
The Netflix recommendation system is definitely not solved yet, but they have had competitions to improve it and you could probably get some good ideas about how to use a user's history to predict their future preferences.