Wikipedia talk:Snuggle/Archives/2013/May

Desirability algorithm

I notice that you say Snuggle will algorithmically determine desirability, using something from STiki. This is very interesting. How is this going to be done? Is it going to be a simple case of looking at the probability of vandalism that STiki assigns to the newcomer's edits? Yaris678 (talk) 21:46, 23 December 2012 (UTC)[reply]

Ahh... Good question. I won't have this ready before the demo, but I hope to have it ready for the full deploy (middle of Jan.). I built a bayesian classifier based on Stiki scores and trained it using manually coded newcomers from this study. It's not great, but it is good enough to support human judgement. Hopefully, I'll get some docs together when I have a chance to start working on this functionality so you can see more about how it works. --EpochFail^(talk|work) 22:32, 2 January 2013 (UTC)[reply]

Sounds interesting. How many new editors were classified in the training set you used? I guess you could use the categorisations given by Snuggle users to create a new training set. Yaris678 (talk) 16:16, 11 January 2013 (UTC)[reply]

This reply is ridiculously late. I'm sorry to have not gotten back to you in so long. For details on the "desirability" metric and how I trained my classifier, see meta:Research:Newcomer_desirability_modeling. --EpochFail^{(talk • work)} 20:07, 30 April 2013 (UTC)[reply]

Thanks. Yaris678 (talk) 17:41, 8 May 2013 (UTC)[reply]

Another new button

We need one more button to classify editors - The three currently do not work.

Golden/Green - Good faith edit, Good edit Yellow - Good faith edit, Bad edit overall (Needs mentoring - Mentors can monitor and take up "grooming" of yellowed users before they get into disputes) Grey - Ambiguous faith edits Red - Bad faith edit

As for the rest, I think you and the current ideas work fine. Still waiting on a different colour coding for article/talk/User talk [Maybe you could use a different colour than Green (Maybe Purple/violet and light purple/violet), it would be less confusing to newer users - Blue isnt used anywhere else so its shades work fine]

Eagerly waiting for more progress!

Cheers, TheOriginalSoni (talk) 22:48, 28 January 2013 (UTC)[reply]

Once I have classified some more users, it seems apparent where the users generally fall. I think under the grey/yellow zone we have mainly these several kinds of users. In all cases, the thing to do is to wait and monitor closely, but who does the monitoring is vastly different for each. Maybe we could implement some kind of strategy to Snuggle-deal with these kind of users.

Possible POV pushers - Newbies editing a single article of their interest, can turn out to be productive/POV pushers (adding unsourced content blah blah blah) - Need to be mentored/politely shown the door
New article creators who get deleted - Need to be escorted to AfC [Now that I see it, deleted articles, and new articles must have something to point them out from the rest of the blocks - Maybe a triangular note, like we have for comments in Excel]
Vandals trolling around - Need to be blocked
Genuine good faith editors with especially bad vocab - Need to be mentored
Editors who dont understand the internal workings of Wikipedia, and get into the Hot Zone with one or more editors - Need to be mentored before things get too hot
Actual grey area candidates.

Those are the main ones I could see. Hope this will come in handy for you and others. Cheers, TheOriginalSoni (talk) 23:17, 28 January 2013 (UTC)[reply]

One more - Automated reverts must be separated from actual ones. If I see a revert, I'll half expect that user to be non-constructive, which wont be the case all the time with ClueBot. TheOriginalSoni (talk) 23:37, 28 January 2013 (UTC)[reply]

I am not sure where to report it, so I'll post it here in the hope that someone will take it to the proper forum (Since Snuggle is how I found it) - Many usernames are Promotional. And when I tried to create a new account to check on why, I realised - Its because the actual policy isnt mentioned clearly enoough. On the create account page, it must be clearly said "Usernames which do not meet our policies can be blocked" or "Your account must be yours, not your organisation's" or "Any username representing an organisation might be blocked per our policies". TheOriginalSoni (talk) 23:48, 28 January 2013 (UTC)[reply]

I share the same concerns as Soni. The good/bad/undetermined split is much too crude. In reality, there is no adequate way of dividing into non-overlapping categories--perhaps no fully adequate way of dividing into fixed categories at all. Attempts to automate or even systematize user advice have rarely been successful here--we either overgeneralize, or we overload with details. The only way that works is fully customized help given carefully by experienced tactful editors, and followed up effectively. We might have enough such editors to do that, but only if everything else stopped here. So I think a realistic system must be built on the same principles as articles tagging: explanations of the usual problems, applied as needed. This would incorporate soni's list of suggestions, but I can think of many others. Perhaps we can develop this, but I think for now, we'd do better with a system with perhaps one or two additional categories: good faith but problematic , and possible bad faith. (I'm trying to think of a synonym for bad faith--its a term that shouldalmost never be used, especially at intial contact.) DGG ( talk ) 15:33, 28 April 2013 (UTC)[reply]

- While I agree that the categories may be limiting, I'm not sure we know what to expand them to yet. I'd like to have a few people use the system first and then re-visit this suggestion once we have a better idea what categories people would like to make use of. --EpochFail^{(talk • work)} 18:49, 30 April 2013 (UTC)[reply]

so use "other" to start off with. DGG ( talk ) 18:53, 15 May 2013 (UTC)[reply]