Or allow page viewers to move something from one column to the other and have the back-end receive notification so that it modifies its sentiment algorithm automagically.

I see some positive ones on the negative side too:

"Batman no!! You can't lose your faith in humanity!"

"You must not lose faith in humanity. [...] - Gandhi"

I don't know how the sentiment analysis is being done (naive Bayes?), but perhaps it could be tweaked to handle this case.

Here is a stab in the dark:

These types of declaratives have two negatives side by side (e.g., "not lose", "can't lose"), making them into a positive. They're somewhat similar to rhetorical questions which have assertions embedded in them in the form of a negative question. From the Wikipedia article, "What have the Romans ever done for us?..." (What...ever...?) "...should be read as The Romans have never done anything for us"... (where a negative is surrounded by a verb).

http://en.wikipedia.org/wiki/Rhetorical_question#Negative_as...

What do people think of this? I'm not entirely familiar with naive Bayes classifier.

Even a tweet about this website itself is categorized as negative: https://twitter.com/LeviDayley/status/313711985437073409

They'll be cancelled out by false positives, eg 'I try to maintain my my faith in humanity. Sometimes I succeed.'

Hi! Creator here, yes the sentiment analysis could definitely use more work but we've found our naiive approach to be about 95% accurate which was good enough for our MVP.

Definite idea, some friends have mentioned it might be useful for brands

Might I ask what exactly is the naive approach? Is it simply searching for keywords like "lose", "lost", etc?

Pretty much: https://github.com/imkevinxu/faithinhumanity/blob/master/fai...