融合词语场景的隐喻情绪识别方法

Metaphorical Emotion Identification by Fusing Word Scene

从认知学角度,隐喻情绪由句子中“源语义场景-目标语义场景”词对的情绪场景融合而成。鉴于此特点,该文提出了融合词语场景的隐喻情绪识别模型。该模型借助情绪词典及大规模语料库,构建了词语情绪分布表示获取算法,用于捕获句子中映射词对的情绪分布表示。在此基础上,利用注意力机制与最大池化策略,编码句子的多情绪场景融合表示,以刻画句子情绪形成的诱因。最后,设计情绪分类器,联合句子情绪及上下文表示作为输入,多角度地构建句子的语义,以提升隐喻情绪识别性能。在隐喻情绪数据集上进行实验,与基线模型和最好评测模型进行对比,该文提出的模型在宏F1值上提升了5.74%与2.73%。另外,定性的实例分析解释了词语场景对隐喻情绪识别的作用。

According to cognitive science,metaphorical emotion is composed of the emotional scenes of the word pair"source semantic scene-target semantic scene"in a sentence.Therefore,this paper proposes a metaphorical emotion identification model by fusing word scenes.This method captures emotion distribution of word pairs with the help of emotion dictionary and large-scale corpus.Then,we use the attention and maxpool mechanisms to encode sentence representations integrated with multiple emotion scenes.Finally,an emotion classifier is designed by combining sen-tence emotion and context representation as input.Experimental results demonstrate the proposed model improves by 5.74%and 2.73%on macro-F value compared with baselines.In additon,qualitative case study explains the role of word scene in metaphorical emotion identification task.