Text Analysts Grapple with Sarcasm

Researchers at The Hebrew University in Jerusalem have devised an algorithm that recognises sarcasm in text - potentially vital to companies analysing online sentiment in social media and other sources.

The team of Oren Tsur, Dmitry Davidov and Ari Rappoport used a number of human volunteers from differing backgrounds to assess around 5,500 sentences from reviews on the Amazon shopping site, and mark them as either sarcastic or non-sarcastic - results were used to build a program able to distinguish between the two. Following further fine-tuning, the program was tested on 66,000 further human-assessed Amazon product reviews and found to successfully pick out sarcastic comments in 77% of cases - impressive considering the difficulty some humans have in spotting it at all.

The authors say that rather than a stock list of phrases, 'a combination of more subtle features' proved the best way to instruct the tool. For example, the book review comment '[I] Love The Cover' might, the report suggests, 'be a genuine complement if it appears in the body of the review. However, recalling the expression 'don't judge a book by its cover' and choosing it as the title of the review reveals its sarcastic nature.'

It goes on to note that products attracting the most sarcastic comments included Dan Brown's novel The Da Vinci Code, and Amazon's Kindle e-reader, suggesting: 'We speculate that one of the strong motivations for the use of sarcasm in online communities is the attempt to 'save' or 'enlighten' the crowds and compensate for undeserved hype.'

The paper, itself ironically titled A Great Catchy Name: Semi-Supervised Recognition of Sarcastic Sentences in Online Product Reviews, and which continues throughout in clear, accessible language (see image above), can be found here.