Also, taking note of how often a user visits (refreshes) the site could be useful (varying the second component) - if I haven't been to the site in a week, it would make sense to show me older stories than if I visited this morning.
1) Spez actually weighs in on such matters
2) the suitability of supervised/unsupervised algorithms for recommending links is discussed (note that Amazon and other highly successful recommendation engines rely upon multi-layered feedback to self-tune, hence my bias towards supervised algorithms)
3) the churn and refresh rate, along with maximization of the desirable turnover and minimization of bombing, is discussed in the context of the 'hot' algorithm and potentially useful changes to the algorithm
You might find it worth your while.