Context-aware emotion cause analysis with multi-attention-based neural network

Abstract Emotion cause analysis has elicited wide interest in both academia and industry, and aims to identify the reasons behind certain emotions expressed in text. Most of the current studies on emotion cause analysis do not consider two types of information: i) the context of the emotional word, which can provide rich emotional details, and ii) the interaction between the candidate clause and the emotional clause (containing the emotional word). The above information is able to provide important clues in emotion cause analysis. In this paper, we propose a multi-attention-based neural network model to address this issue. First, our model encodes the clause via bidirectional long short-term memory, which can incorporate the contextual information of the word. Second, a multi-attention mechanism is designed to capture the mutual influences between the emotion clause and each candidate clause, and then generate the representations for the above two clauses separately. With this design, our model creates better-distributed representations of the emotion expressions and clauses. Finally, these representations are fed into a convolutional neural network to model the emotion cause clause. The experimental results show that our proposed approach outperforms the state-of-the-art baseline methods by a significant margin.