U of G research aims at interpreting sarcasm

Sarcasm is a way to convey meaning in language to poke fun at what someone is trying to say and at the University of Guelph they have developed a model to take sarcasm in social media one step further. 

Fattane Zarrinkalam, a professor in the school of engineering along with researchers at Ferdowsi University of Mashhad in Iran, have helped develop the model so instead of sarcasm detection, the model uses sarcasm interpretation.

Most of the current work is on sarcasm detection because sarcasm interpretation is a lot more complicated, said Zarrinkalam.

“In social media, users publish sarcastic posts to express and you know, unpleasant subject or insult or criticize some conditions,” she said.

She gave an example of different cell phone brands, and there are sarcastic reviews posted, as a result of this, a cell phone brand that isn’t very good, could be rated higher.

Social media mining is a tool used in marketing with recommender programs, to recommend products to users. In some cases positive phrases like, “such a great phone” have positive and negative meanings, said Zarrinkalam.

“The natural language processing system, consider it as a positive message and probably recommend this phone to you, which is not correct,” she said.

“It's really necessary to detect sarcastic text on social media. And then actually detecting sarcastic text is important, but the more important thing is to convert it, to convert it to the intended meaning and then use it in social media.”

To do this, the researchers used deep neural networks, a type of machine learning which is supervised.

Taking the research into the future means adding context to the social media posts, to model the user, their personality, location, job, and nationality, said Zarrinkalam.

“We are just using the relation between words in the text or sentiment of the words. And you're putting attention … on different words,” she said.

Zarrinkalam has experience with model users with her PhD thesis and her experience analyzing content of user activities on social media will go into the next two months of work on integrating it into deep learning models.

Sarcasm can be used in a partial sense by analyzing social media posts for it, and why they are changing brands of a product.

It will help “improve their accuracy and the decisions that they are going to make based on their results,” said Zarrinkalam.