Researchers at Carnegie Mellon University's School of Computer Science created a system that will allow computers to recognize instances of sarcasm on Twitter.
Part of their research include using the hashtag "#sarcasm" on tweets and then testing it when new messages are sent.
Previously, the group would identify sarcasm in text using linguistic cues. In the latest experiment, the team included new contextual bases such as the author of the text and the content of the tweet.
"We present here a series of experiments to discern the effect of extra-linguistic information on the detection of sarcasm, reasoning about features derived not only from the local context of the message itself (as in past work), but also using information about the author, their relationship to their audience and the immediate communicative context they both share," said the researchers.
According to the paper, the researchers discerned whether a tweet is sarcastic or not by adopting the so-called "binary logistic regression," along with cross-validation and split on authors (making sure that tweets by a similar author do not appear in multiple splits).
The group used four types of features in their models. These include the Tweet Features (covers the immediate tweet that is being predicted); the Author Features (includes the author of the tweet and the author's historical data); the Audience Features (the addressee of the tweet, historical data of the addressee, and the history of his interaction with the author); and Response Features (the interaction between the predicted tweet and the tweet which it is addressing to).
"We find that the strongest audience-based features that act as markers of sarcasm in this dataset are not those that suggest intimacy between the author and audience; the strongest audience predictors of sarcasm are the absence of mutual mentions (at least one mutual mention is a contraindicator, and at least two is more so); living in different time zones (not being geographically proximate) and features of celebrity (being verified and having many followers)," wrote the researchers.
By detecting sarcasm in social media posts, various efforts on tracking and filtering certain types of language on the web can be done more efficiently. Computers will then be able to differentiate a statement intended to be said as a joke and one intended to convey a serious message. Eventually, computers will also be able to determine the presence of rapport and context found in a pool of users.
Lastly, detecting sarcasm more reliably can help in filtering undesirable content from Web trolls or from posts that are found in social media. In the future, computers will be able to recognize humorous images and may even have a way to create their own.
