2 ms·
"The hard part is getting the initial corpora of stupid and non-stupid text" Can't you simply use comment votes for this - use the comments with the least vote
by mixmax 19y ago
"The hard part is getting the initial corpora of stupid and non-stupid text"
Can't you simply use comment votes for this - use the comments with the least votes? According to the users of a site these would be the stupid ones. Maybe comments with some algorithmic tool to include the karma of the user that wrote the comment.
The advantage of this approach is that it will work across sites where the definition of stupid might differ.
- comatose_kid 19y agoI don't know if the stupidity of the message correlates to its vote tally. For example, a comment with a negative rating may hold an unpopular but valid view. In fact, if the mapping was that good, one wouldn't need to run a 'stupid filter' on the message body in the first place.
- derefr 19y agoThe data might be more useful together--first generate a simple intelligence-scale number (between -1 and 1), then weight all of a user's votes by their intelligence, and recalculate everyone's intelligence from their weighted karma. That is to say, if a lot of stupid people hate you, you appear smarter. The problem might be with trolls, where the stupid are tricked and vote down in retribution, but the intelligent notice right away but are simply amused at the skill of the execution, and vote up in humor. This makes the troll appear much more intelligent than they are.