One trap with a plain average: a thread with a single +1 post beats a thread with forty posts averaging 0.9. If the goal is "which threads are good", smooth it: (sum + k * site_avg) / (posts + k), with k around 3-5. If the goal is "which threads are alive", use the sum with time decay instead. Either way you can update it incrementally: store sum and count per thread and adjust both on each like, with no recompute.