Do you know of any extensions that put googly eyes over spiders? I really want this, but ironically I am too squicked out by spiders to go acquire a training set and tweak it myself.
It is also entirely possible that since I have only posted this to HN and Facebook thus far, it has not yet reached its ultimate audience of idiot cyberbully trolls :(
Ok, seems like it's doing pretty well in favor of positivity. I searched some common nice/nasty words in the DB and I'm seeing on the order of:
- 25-70 hits for strings containing words like nice/fun/smart/kind
- 1-3 for dumb/ugly/stupid/boring/fag
- no incidences of slut/bitch/hate/crap/... (I searched a LOT)
- one racial slur that I could find
- surprisingly no negative uses of curse words (one "fucking awesome")
This is a way lower incidence of negativity than anticipated. Due to the simplicity of the text inputs, I'm going to hold off on preemptive moderation that would change the feel of the app and see how it works if I just flag blacklisted words.
Haha, that's why I don't display numbers on the bar charts. They're just to get the most important characteristics to bubble to the top. I kept the overall aggregate counts because seeing those large numbers seems to give people a nice quick shot of good feels. What do you think about those?
Btw, many thanks for pointing this case out. I really don't have good ideas on how to prevent this besides going the brute force route of flagging negative words.
Ahh I see, fishing for compliments is definitely a bad use case that's hard to circumvent. Any ideas? I attempt to subtly discourage narcissistic submissions in the prompts but people will probably still do it.
Clarification of analogy: if a doodle link gets posted to an internet forum (as happened with my example page), it will get spammed, but trollbait pages are irrelevant -- pages which are generated for their intended purpose do not run into that problem by virtue of obscurity.
Yup. I deliberately didn't implement any authentication because I wanted it to operate more like a Doodle -- a link gets generated for a purpose of no particular celebrity and sent to respondents for low-threshold participation. I think I will lock down submissions on the example page to make this clearer.
It's pretty fun for users who are using it for its intended purpose, and not very trollable if you use the default URL randomization.
Also bear in mind I made this in a day. It might be neat to add sentiment analysis / negative word flagging in order to weed out all-troll submissions.
Personally I'm surprised by the extremely high disapproval/approval ratio for C++. It does what you want it to, it's performant, the syntax doesn't make you jump through unnecessary hoops, and it isn't full of warty language design problems like Javascript. I may be too biased to see the issues because it was my first language. A C++ hater's perspective would be appreciated.