@Glorfindel I really love your auto edits, seeing your second edit in jasper-reports... I was what the heck are you doing in jasper-reports.. but yeah I found the git page!
@Machavity hmm ok, but I dont agree. Not even sure it should be deleted, maybe edit (note user even edit by themself before deletion to remove link). Good faith rule
@Undo I just feel that your old teaching was to be really careful about being overzealous about flagging. So I'm a bit lost, already the previous answer it's enough with old rulez to NAA (a non working link on a norton question, red flag?)
@Mr.YouKnowWhoIAm That lib is an api, hence you call it we web with whatever language (get method returning main result), currently we don't have a public endpoint, but you can setup a free tier Amazon lambda and use it
As for ML algos I have tested a lot of different ones, the best result I think is with naive bayes, using 2-3 ngrams, a stop word list, and a good stemmer.
@quartata yeah, on vacation so only with mobile phone. The current feed used is about 3000 bad and 3000 good. The 50000k is not labelled data, I don't think you can run automatic stuff to label it, having a good feed is fundamental in machine learning, it's the hard part :)
I'm in vacation now but if you like to have a MySql dump of 50.000 comments that have been classfied also with low score let me know, I will try to connect to our server on off these days and male a dump
@thesecretmaster Hiya we got tons of other data also (more then 50.000 conments that have been classfied bad) and we have reclassfied all the SE data given in meta post, its fairly poor data. Furthmore I have already ask SE Joe, if they will make thier data aviable check Joe's latest blog post,
