Posted Nov 7, 2012 11:27 UTC (Wed) by farnz (subscriber, #17727)
The trouble is that I want a killfile based on content, not on who's commenting - if Corbet and company have an algorithm that can accurately judge whether a particular comment thread is interesting to me or not, then they've got something genuinely worthy of a software patent (given the state of the art in predictive technology today). I don't want to miss out when someone's saying something topical and interesting, even if most of their output is off-topic spouting; as the expense of filtering through these big comment threads is high, the value of an LWN subscription is reduced.
Posted Nov 7, 2012 12:06 UTC (Wed) by man_ls (guest, #15091)
Actually this long off-topic thread has served to feed my killfile with at least two more usernames, so right now I see only 28 comments, all on-topic. Get trolled once, do not get trolled twice.
Posted Nov 7, 2012 14:31 UTC (Wed) by nix (subscriber, #2304)
(Since Gnus is an Emacs-specific program, and Emacs was originally written by RMS and was the first program ever to be GPLed, this gets a bit less off topic. Not that RMS uses Gnus. No, no, he is one of the three people on the planet to still use rmail, and IIRC still uses a web-by-email robot. Whenever I think I hate change in my workflow to a pathological degree, I just think of RMS and am happy again. If RMS can function with *that* degree of workflow-conservatism, mine is clearly not too bad.)
Posted Nov 7, 2012 16:06 UTC (Wed) by man_ls (guest, #15091)
[Ironically these last two super-off-topic comments take the thread closer to on-topic; I guess we have reached the limits of topicland and went back, it must be a round world.]
Posted Nov 7, 2012 16:03 UTC (Wed) by pboddie (guest, #50784)
if Corbet and company have an algorithm that can accurately judge whether a particular comment thread is interesting to me or not, then they've got something genuinely worthy of a software patent
I imagine that a Bayesian classifier as seen in things like Spambayes would do the trick, one again showing that something appearing patentable to one observer is most likely to be familiar and even mundane to another.
Posted Nov 7, 2012 20:09 UTC (Wed) by farnz (subscriber, #17727)
A Bayesian classifier is not enough - to meet the stated goal, the classifier has to identify cases where due to circumstances outside its knowledge (e.g. me reading LWN while sitting up with a sick child), my interests are broader than normal, and identify cases where my tolerance for off-topic is reduced (e.g. because said sick child has resulted in sleep deprivation). An algorithm that could detect my mood based purely on my reading patterns, and tweak a Bayesian classifier accordingly would be quite impressive.
Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds