To the best of our knowledge, no text analytics software currently exists that is capable of high-quality sarcasm detection.
The fundamental difficulty with sarcasm is that, by design, the words are being used to convey the opposite of their literal meaning, and it is left to the listener or reader to determine this fact. To aid the listener, a speaker uses tone of voice and body language, which are missing entirely from text.
A speaker or writer can also rely on the listener or reader to know their actual opinions -- if Pat knows that Sandy hates spinach, and Pat gets a text message from Sandy saying “My mother made spinach for dinner. I’m so excited! That’s my favorite!”, then Pat can be pretty certain that Sandy is being sarcastic. But a computer, like any other stranger, won’t have that information and is likely to miss the sarcasm.
Thus far, we have found sarcasm to be a statistically insignificant problem, as the overwhelming majority of text is sincere. However, if you have a data set containing an unusually high level of sarcasm, you may want to use another data set.