whathappenedto also suggested the article you linked, in this reply: https://news.ycombinator.com/item?id=5430495 https://news.ycombinator.com/item?id=5430495 (

whathappenedto also suggested the article you linked, in this reply: https://news.ycombinator.com/item?id=5430495 https://news.ycombinator.com/item?id=5430495 (the "Dynamic bayesian network model" link). The approximative solution at the end of the article seems really interesting :-) It'd feel better to implement something that is theoretically sound. I suppose the algorithm could need some tweaking, since a discussion is a tree or a graph, but a search result listing is... a list. Generating a slightly different order for every reader also seems like a good idea, and fairly easy to implement. — So as not to make people confused when they reload the page (by shuffling comments around), perhaps one could use the user ID or IP number as random seed. (May I ask, how come you know how search engines solve this problem?)