基于跨语言知识增强和对比学习的隐式情感分析OA
Implicit sentiment analysis based on cross-lingual knowledge enhancement and contrastive learning
隐式情感分析是情感分析领域的一个重要类型,对于准确理解文本中隐含的情感信息至关重要.然而,现有的情感分析方法,通常只关注显示情感,忽略了隐式情感在实际评论中的重要作用.特别是在中文语境下,因为语义丰富且常使用隐喻、比喻等修辞手法,隐式情感更加难以捕捉和分析.针对该问题,提出了一种基于跨语言知识增强和对比学习的解决方法.首先,利用预训练模型注入跨语言知识进行知识增强,通过引入跨语言知识来丰富语料库,提高模型对语义的理解能力.其次,利用双向长短时记忆网络(BiLSTM)学习句子的正反向语义信息,从而更好地捕捉句子中的上下文特征.最后,采用对比学习的方法获得基于知识和融合上下文特征的句子表示.实验结果显示,所提方法在多个数据集上都获得了最优性能,验证了方法的有效性.实验结果表明,该方法在隐式情感分析任务上取得了显著的提升.
Implicit sentiment analysis plays a crucial role in understanding latent emotional expressions embedded in textual content.However,most existing sentiment analysis methods primarily focus on explicit sentiment cues while neglecting implicit sentiment information commonly found in real-world user-generated texts.This challenge is particularly pronounced in Chinese texts,where rich semantic structures and rhetorical devices,such as metaphors and similes,frequently obscure underlying sentiment expressions.To address these issues,a novel implicit sentiment analysis framework,termed CKECL,is proposed.First,a pre-trained translation model is employed to generate cross-lingual knowledge for data augmentation.The generated cross-lingual information enriches se-mantic representations and improves the model's capability to capture implicit sentiment patterns.Second,a bidirectional long short-term memory(BiLSTM)network is adopted to model contextual dependencies from both forward and backward directions,thereby enhancing sentence-level semantic representation.Finally,information from sentences,is used to better capture contextual features within the sentences.Finally,contrastive learning is introduced to learn more discriminative sentence representations by integrating cross-lingual knowledge with contextual features.Experimental results on multiple benchmark datasets demonstrate that the pro-posed approach consistently outperforms competing baselines,confirming its effectiveness.
方万红;林煜明
桂林电子科技大学 广西可信软件重点实验室,广西 桂林 541004桂林电子科技大学 广西可信软件重点实验室,广西 桂林 541004
信息技术与安全科学
隐式情感情感分析预训练模型知识增强双向长短时记忆网络对比学习
implicit sentiment analysissentiment analysispre-trained modelknowledge enhancementbidirectional long short-term memory networkscontrastive learning
《桂林电子科技大学学报》 2026 (4)
394-402,9
国家自然科学基金(62062027)桂林电子科技大学研究生教育创新计划(2022YCXS093)
评论