Detecting the target of sarcasm is hard: Really??

Highlights：

• What makes this task challenging is that there can be multiple targets or no targets at all in a sentence. Additionally, what constitutes sarcasm can be very much subjective.

• We first built a classifier that distinguishes the presence of target and then we employ deep learning methods that are used in Aspect-Based Sentiment Analysis to extract the target.

• The key takeaway message is that detecting the target accurately is very hard. The lack of consistency among human annotators is hindering creation of a high accuracy classifier.

• We also suggest some ways of addressing the gaps such as having a convention for constructing data sets that would help to reduce human biases when judging them.

摘要

•It is important to identify the target of sarcasm in combating cyberbullying or in ecommerce for retailers to know if their products are being ridiculed. Research in this area is still in its infancy.•What makes this task challenging is that there can be multiple targets or no targets at all in a sentence. Additionally, what constitutes sarcasm can be very much subjective.•We first built a classifier that distinguishes the presence of target and then we employ deep learning methods that are used in Aspect-Based Sentiment Analysis to extract the target.•The key takeaway message is that detecting the target accurately is very hard. The lack of consistency among human annotators is hindering creation of a high accuracy classifier.•We also suggest some ways of addressing the gaps such as having a convention for constructing data sets that would help to reduce human biases when judging them.