融合VAD知识的情感分布增强细粒度情绪识别方法
打开文本图片集
中图分类号：TP39 文献标志码：A 文章编号：0253-2395（2025）04-0666-11
Abstract：Fine-grainedemotionrecognition models，whichemploydozensemotioncategories tomodelhumanemotions，arecapablecaptringsubtleemotionalexpresions moreaccuratelythantraditionalmodels.However，existingemotionpredictionodelshavenotfullyconsideredthecomplexcorelationsthatexistamongthenumerousfine-grainedemotions.Toadresthisissue， thispaperproposesanVAD（Valence-Arousal-Dominance）EmotionDistributionAugmentedFine-grainedEmotionRecognition （EDAER）.EDAER models the emotional correlations in the VAD space using emotion distributions andcombines textualsemantic informationwithpsychologicalpriorsforfine-graiedemotionrecognition.TheEDAERmodelconsists treemodules：asemanticinformationmodule，anemotiondistributioninformationmodule，andafusionpredictionmodule.Thesemanticinfrmatiomodulextracts textual semanticfeaturesusingapre-trained BERT（Bidirectional Encoder Representationfrom Transformers）model; theemotiondistribution information module generates emotion distributionsforemotion words basedon VADdistancemetrics to measuretesimilaritybetweenemotions;andthefusionpredictionmodule integrates textualsemanticfeaturesandemotiondistributioninformationthroughanatentionmechanism topredictemotions.ExperimentalresultsontheGoEmotionsdatasetdemonstrate that the macro-average F1 score the EDAER model reaches 51.75% ，outperforming both theKEA（Knowledge-Embedded Attention）model，whichsesemotionlexiconsasexteralknowledge，andtheHGCN-EC（HierarchyGraphConvolutionNetworksbased EmotionRecognition）model，whichutilizesherarchcalemotionrelatioshipsasextealknowledge.Notablyfortheotion categories with fewer samples，EDAER significantly outperforms other models in terms F1 score.These results validate that modelingemotionalcorelationsintheVADspacethroughemotiondistributionscanefectivelycapture knowledgerelated torareemotions，thus improving the model's ability to recognize fine-grained emotions.
Key words：VAD emotion space; external knowledge; emotion classfication; GoEmotions
0 引言
随着自媒体时代到来，文本情绪识别成为自然语言处理领域的重要研究方向，在医疗健康，网络舆情，消费者行为分析等方面得到广泛应用[1-3]。（剩余16033字）
