A Label Similarity Attention Mechanism for Multi-label Emotion Recognition

Abstract

In recent years, multi-label emotion recognition has attracted more and more attention, and its purpose is to identify the emotional state of a text, such as happiness and like. At present, most of the methods to solve the problem of multi-label emotion recognition focus on analyzing the semantics of the text, ignoring the semantic information brought by emotion tags, and not modeling the semantic relationship between emotion labels and text. In this paper, a method of multi-label emotion recognition based on label similarity attention mechanism is proposed. In this method, bidirectional LSTM (Long Short-Term Memory) is used to obtain the context knowledge of the text, so as to achieve the purpose of learning the global information of the text. The label features are learned by the label encoder. The label similarity attention mechanism is used to obtain the correlation matrix between the label and the word representation, and further strengthen the semantic information. At the same time, CNN (Convolutional Neural Networks) is used to obtain local semantic information. The semantic representation of sentences is enriched. Finally, emotion recognition is carried out through the classifier. In the classifier, this paper proposes a joint loss function, which enables the model to learn the emotional characteristics of texts and labels at the same time. The model combines the global and local information of the text to show the semantic relationship between the text and the emotional label. The experimental results on the NLPCC2014Task1 dataset show that the proposed model has a certain performance improvement compared with the benchmark method.