We've also done some clustering of ASF Archives (See MAHOUT-588) that could show similar items, potentially. I've often thought it would be cool to automatically classify email as to the expertise level required (which could also be done w/ Mahout) and I've always wanted our Mailing List manager to do a search of the archives first, before actually forwarding the email to the user list. If the search turns up answers, send them back to the user and ask them if they answer the question or not. If not, then the mail can go through.
Just fun stuff to think about how to help people find answers better... -Grant On May 14, 2011, at 1:45 PM, Ted Dunning wrote: > A student popped up a while ago on the Mahout mailing with a very nice > little magic program that would sift through email archives to find good > question/answer pairs in email threads. > > The results were quite impressively good. The program didn't find a lot of > pairs, but the pairs it did find were uniformly pretty excellent. > > Maybe a secondary search index based on the output of such a program would > be useful. > > On Sat, May 14, 2011 at 3:20 AM, Ross Gardler <rgard...@apache.org> wrote: > >> Sent from my mobile device (so please excuse typos) >> >> On 13 May 2011, at 02:31, David Blevins <david.blev...@gmail.com> wrote: >> >>> For me tagging and voting and (i forgot) the marking the question >> answered (thanks, Benson) are the parts I would love. >>> >>> I write some really good responses sometimes and even *I* have a hard >> time finding some of my old responses in the list archive haystack. >> >> Right. I always ask users to provide a patch if they find an answer in the >> mailing list useful. Of course it rarely happens (even with devs). >> >> Keeping things simple, could we provide a feature in the CMS that simply >> copies a mail from our archives (with backlinks) into the CMS system for the >> appropriate project? >> >> A link to this could also be provided in the footer of each mail (only >> works for committers). >> >> In the CMS we could have some magic system to build an index. >> >> I appreciate this has now moved away from stack overflow (I changed the >> subject) but for any Perl hackers looking for something useful to do on a >> weekend I would certainly use such a feature. >> >> This could grow to fancy tagging, tracking and more. But I believe thus is >> a reasonably simple thin to do that would provide immediate benefits. >> >> Ross >> >>> >>> And to avoid the "tag names can be spam" issue having so that only >> committers can introduce new tags would be fine for me. It could be a file >> in svn or something else equally lame but functional. >>> >>> >>> -David >>> >>> On May 12, 2011, at 6:04 PM, Ted Dunning wrote: >>> >>>> There is another factor that comes into play. QA sites like SO also >> blend >>>> in wiki and trust mechanisms. Thus, highly rated users can and do >> rewrite >>>> questions to be more answerable/understandable. They can also rewrite >>>> answers if necessary. >>>> >>>> Without automated karma, the moderation function has to be granted >> manually >>>> which is a process that doesn't scale as easily and is subject to attack >> by >>>> cabals. That way lies wikipedia's dictatorship of the editor >> proletariat >>>> and associated drop in user participation. That is fine for a largely >>>> static knowledge base, but SO addresses much more dynamic topics in a >> way >>>> that engages the readership much more strongly. Moreover, the feedback >>>> cycle essentially guarantees that the moderators reflect the interests >> of >>>> the voting public. >>>> >>>> On Thu, May 12, 2011 at 5:47 PM, David Blevins <david.blev...@gmail.com >>> wrote: >>>> >>>>> Another thought. Sometimes I wonder how hard it would be to just allow >>>>> tagging and voting on top of a plain mailing list emails. A simple DB >> with >>>>> the messageId as the key for tags and vote count then a slightly >> fancier >>>>> archive view than we have now. And hey, markdown happens to look nice >> as >>>>> plain email. I've actually been indenting code snippets for years. >>>>> >>>>> I admit I like getting SO points and badges but they do not factor in >> at >>>>> all when looking for the right answer. >>>>> >>> >> -------------------------- Grant Ingersoll Lucene Revolution -- Lucene and Solr User Conference May 25-26 in San Francisco www.lucenerevolution.org
