Abstract: Emotion distribution learning（EDL） is a recently proposed effective model for multi-emotion analysis. The main idea is to record the expression degree on each emotion of examples by emotion distribution, which is appropriate for dealing with emotion analysis tasks with emotional ambiguity. An emotion wheel attention based emotion distribution learning（EWA-EDL） model is proposed to address the problem that existing EDL methods rarely consider the priori knowledge of emotional psychology. The EWA-EDL model generates a prior emotion distribution for each basic emotion describing the emotion psychological relevance, and incorporates the prior knowledge based on the emotion wheel into the deep neural network through an attention mechanism. The EWA-EDL model uses an end-to-end approach to train the deep network to learn both emotion distribution prediction and emotion classification tasks. The EWA-EDL model consists of five main components, namely, input layer, convolutional layer, pooling layer, attention layer, and multi-task loss layer. Comparative experiments on eight commonly used textual emotion datasets show that the EWA-EDL model outperforms comparative EDL methods on both emotion distribution prediction and emotion classification tasks.

Key words: emotion distribution learning, emotion wheel, attention mechanism, emotion classification

摘要： 情感分布学习是一种近年提出的有效的多情绪分析模型，其核心思路是通过情感分布记录示例在各个情绪上的表达程度，适于处理存在情绪模糊性的情感分析任务。针对现有的情感分布学习方法较少考虑情感心理学先验知识的问题，提出一种基于情感轮注意力的情感分布学习（emotion wheel attention based emotion distribution learning，EWA-EDL）模型。EWA-EDL模型为每种基本情绪生成一个描述情绪心理学相关性的先验情感分布，再通过注意力机制将基于情感轮的先验知识直接融入深度神经网络。EWA-EDL模型采用端到端的方式对深度网络进行训练，同时学习情感分布预测和情绪分类任务。EWA-EDL模型主要由5部分构成，分别为输入层、卷积层、池化层、注意力层和多任务损失层。在8个常用的文本情感数据集上的对比实验表明，EWA-EDL模型在情感分布预测和情绪分类任务上的性能均优于对比的情感分布学习方法。

关键词: 情感分布学习, 情感轮, 注意力机制, 情绪分类