Two-stage supervised ranking for emotion cause extraction

Abstract

Abstract Textual emotion analysis is a challenging research topic in the field of natural language processing (NLP), which plays an important role in related NLP tasks, such as opinion mining and personalized recommendation. Existing research on emotion analysis has focused mostly on detecting types of emotions, and has solved problems using classification-based methods. Recently, fine-grained emotion analysis has attracted the attention of researchers for probing the essential elements of emotions, such as the causes, experiencers and results of emotion events, which could help further elucidate textual emotions in more depth. In this paper, we focus on the task of emotion cause extraction, aiming to recognize the causes in sentences that provoke certain emotions. We propose a two-stage supervised ranking method for accurately extracting the emotion causes based on information retrieval techniques. In the first stage, we measure the complexity of provoked emotions using query performance predictors to distinguish the number of causes for each emotion in contexts. In the second stage, we incorporate the emotion complexity into learning an autoencoder-enhanced ranking model for accurately extracting the causal clauses. We also extract abundant emotion-level clause features for clause representations as the learning samples. We evaluate the proposed method on an existing dataset for emotion cause extraction and demonstrate that our method significantly outperforms the state-of-the-art baseline methods. The proposed method is effective in extracting textual emotion causes in sentences, which can greatly benefit in-depth emotion analysis for effective cognitive computing.