I'm just sooo happy to be sitting here reading through an eight-page PDF on algorithms. Seriously. Nothing in this world makes me happier than poring over phrases like "detailed results of the 5-fold cross validation of various components of the algorithm are summarized in Table 2."
If a new sarcasm-detecting algorithm out of Jerusalem's Hebrew University really knows what it's doing, it should be able to tell that I was just kidding there. Yeah, right. No, actually I was.
After an exhaustive look at word, syntax, and punctuation patterns in written user-generated content, the researchers came up with SASI (PDF), or Semi-supervised Algorithm for Sarcasm Identification, which can recognize sarcasm in online sentences and assign each sentence to a sarcastic class (not all sarcasm is created equal, of course). Meager attempts at sarcasm aside, this is a pretty novel idea that could possibly aid those pure souls who lack a sarcasm and irony meter--and could even have commercial applications.
One idea here is that automated sarcasm recognition could help improve review summarization and opinion-mining systems, since the inherently subtle and ambiguous nature of sarcasm sometimes makes it hard even for humans to decide whether a comment is sarcastic. According to the researchers--Oren Tsur, Dmitry Davidov, and Ari Rappoport--studies of user preferences suggest some consumers find sarcastic reviews biased and less helpful.
The Hebrew University team--which will present its findings next week at the International Conference for Weblogs and Social Media in Washington, D.C.--closely examined some 66,000 Amazon reviews for 120 products including books, music players, digital cameras, camcorders, GPS devices, e-readers, game consoles, and mobile phones.
Identifying cues common to sarcasm in online communication (excessive use of capital letters as in: "Well you know what happened. ALMOST NOTHING HAPPENED!!!"; puns; and explicit contradictions), the researchers created a complex algorithm in which a small number of sarcastic sentences "teach" the software to recognize sarcasm. They say the software precisely identifies sarcastic sentences 77 percent of the time--no small feat given the elusive nature of sarcasm, its intractable relationship to cultural context, and differences between the spoken and written varieties. … Read more