Sarcasm is difficult to detect online, whether as a user or as an algorithm. Sentiment analysis can help, but there are limits to its effectiveness. An article from cloud-based social intelligence agency Infegy examines the problems resulting from the use of sarcasm online and offers solutions for more accurately identifying and dealing with sarcasm.
- People can’t reliably interpret sarcasm. Sarcasm can be quite nuanced, especially in text-only mediums. Users are notoriously bad at both identifying and creating sarcasm, according to a study cited by Infegy. Given that users have a hard time, it’s not hard to imagine marketers struggling with it, too. Even academic studies have attempted to automatically detect sarcasm, and one achieved a success rate of 80 percent, but only when the data is carefully selected to include a large sample of tweets hashtagged #sarcasm.
- Sarcasm varies and requires context. Most sarcasm isn’t obvious or tagged #sarcasm. Algorithms would need better behavioral analysis capabilities if they are to identify sarcastic statements. And teaching an algorithm context is no easy task. This may cause issues with customer-service requests on social media, so be wary of trick statements.
Training sets for machine learning are limited and can be overfit.
Algorithms can get hung up on the examples provided, which results in t
he algorithm becoming overfitted. In other words, the algorithm only recognizes only similar examples, and ultimately misses out on context and nuance. With context in mind, David Bamman and Noah A. Smith from the school of computer science at Carnegie Mellon University developed a context-sensitive method for determining sarcasm on social media. In addition to the text of the tweet, their model takes into account the author, the intended recipients and responses to the original tweet.
While algorithms can be refined, there is still a long way to go. When social networks have social bots and other hurdles, don’t leave all your analysis to algorithms; get your teams involved.
Image courtesy of Shutterstock.