Multi-label text emotion classification based on self-attention mechanism and emotion semantic similarity modeling

Subjects: Library Science，Information Science >> Library Science submitted time 2024-05-04

YINGYING NI

Abstract： 【 Objective 】 Aiming at the problem that existing multi-label text emotion classification algorithms are difficult to model and utilize the semantic correlation between emotions, a multi-label text emotion classification method based on self-attention mechanism is proposed. 【 Methods 】 In this paper, a multi-label text emotion classification neural network (AttEmoNet) based on self-attentional emotion association modeling was proposed. The semantic meaning and similarity of emotion labels were modeled by self-attentional emotion association module. A text encoder based on large-scale pre-training model was used to encode the input text as a semantic vector. Finally, the matching degree of text semantic vector and emotion semantic vector is calculated by neural network to realize more accurate emotion category recognition. 【 Results 】 The validity of AttEmoNet was verified by comparison experiments on NLPCC2014 and GoEmotions. The results show that the text sentiment classification performance of AttEmoNet is significantly improved compared with the baseline methods Random, cnsenti, SVM and BERT. Compared with the existing optimal baseline method, the classification Precision of AttEmoNet increased by 13.33% at the highest, the Recall by 21.80% at the highest, and the F1-score by 12.74%. Meanwhile, the emotional semantic association matrix modeled by the AttEmoNet has good interpretability, which proves that the ATTEMONet has a good ability to model emotional semantics. Limitation The emergence of large language model opens up a new idea for multi-label text emotion classification. In the future, the advantages of AttEmoNet and large language model will be combined to realize a more accurate and efficient multi-label text emotion classification algorithm. Conclusion In this paper, a multi-label text emotion classification neural network based on self-attention mechanism for emotion correlation modeling is proposed, which improves the emotion semantic and correlation modeling ability of the text emotion model and the performance of emotion recognition. The validity of the study is verified by comparative experiments on two public data sets.

Peer Review Status:Awaiting Review

 Hits 476  Downloads 117  Comment 0