CAN-GRU: a Hierarchical Model for Emotion Recognition in Dialogue

CAN-GRU: a Hierarchical Model for Emotion Recognition in Dialogue Ting Jiang Bing Xu Tiejun Zhao Sheng Li Laboratory of Machine Intelligence and Translation School of Computer Science and Technology, Harbin Institute of Technology Harbin, China jiangting hit@163.com {hitxb,tjzhao,lisheng}@hit.edu.cn Abstract Emotion recognition in dialogue systems has gained attention in the field of natural language processing recent years, because it can be applied in opinion mining from public conversational data on social media. In this paper, we propose a hierarchical model to recognize emotions in the dialogue. In the first layer, in order to extract textual features of utterances, we propose a convolutional self-attention network(CAN). Convolution is used to capture n-gram information and attention mechanism is used to obtain the relevant semantic information among words in the utterance. In the second layer, a GRU-based network helps to capture contextual information in the conversation. Furthermore, we discuss the effects of unidirectional and bidirectional net- works. We conduct experiments on Friends dataset and EmotionPush dataset. The results show that our proposed model(CAN-GRU) and its variants achieve better performance than baselines. 1 Introduction As an important component of human intelligence, emotional intelligence is defined as the ability to perceive, integrate, understand and regulate emotions(Mayer et al., 2008). Emotion is the essential dif- ference between human and machine, so emotion understanding is an important research direction of artificial intelligence. As the most common way for people to communicate in daily life, dialogue con- tains a wealth of emotions. Recognising the emotions in the conversation is of great significance in intelligent customer service, medical systems, education systems and other aspects. According to (Poria et al., 2015), textual features usually contain more emotional information than video or audio features, so we focus on the emotion analysis of dialogue text and aims to recognize the emotion of each utterrance in dialogues. There are some challenges in this task. First, the length of an utterance may be too long, making it difficult to capture contextual information. Furthermore, a dialogue usually contains lots of utterances, therefore, it’s hard to grasp long-term contextual relations between utterances. Second, the same word may express different emotions in different contexts. For example, in Table 1, while in different dia- logues, the word ‘Yeah’ can express three different emotions, that is , joy, neutral and suprise. To tackle these challenges, we propose a hierarchical model based on convolutional attention network and gated recurrent unit (CAN-GRU). Existing works pay little attention to the extraction of semantic information within an utterance. In this work, we focus on this problem. Our proposed model can extract n-gram information by CNNs and use self-attention to capture contextual information within an utterance in the first layer. Moreover, we utilize a GRU-based network to model the sequence of utterances in the second layer, which can fully combine the context when analyzing utterance emotion and solve the problem of long-term dependence between texts at the same time. 2 Related Work Text emotion recognition is one of the most hot topic in natural language processing. …

Content truncated to 500 words