面向社交媒体的反讽识别

Social media-oriented sarcasm detection

反讽是社交媒体中常用的一种修辞方法,反讽的存在对传统的情感分析或观点挖掘带来了挑战。反讽修辞中一种常用的表达形式为使用极性相反的情感词来表达"前后情感矛盾"。本文针对该形式的反讽,提出了一种基于注意力机制的神经网络模型,该模型可以捕捉一句话中的前后情感矛盾的两个词从而推断是否为反讽。该模型不考虑句子的上下文,仅从句子本身的结构出发,计算任意两个词之间的注意力分数从而发现导致反讽的关键词。本模型在多个数据集上取到了很好的效果,并且该模型有较好的可解释性。

Sarcasm is a rhetorical method commonly used in social media.The existence of sarcasm poses a challenge to traditional sentiment analysis or opinion mining.A common form of expression in sarcasm rhetoric is the use of emotional words with opposite polarities to express"inconsistency."In this paper,a neural network model based on attention mechanism is proposed for this form of sarcasm.This model can capture the two inconsistent words in a sentence to infer whether it is sarcasm.The model does not consider the context of the sentence,only from the structure of the sentence itself,calculates the attention score between any two words to find the keyword that leads to sarcasm.This model has achieved good results on multiple data sets,and has good interpretability.