基于深度学习的藏文讽刺对话识别研究

Research on Tibetan Satirical Dialogue Recognition Based on Deep Learning

针对情感分类研究中讽刺表达检测困难,还未有学者对藏文讽刺识别展开研究的问题,本文创建了藏文讽刺对话数据集,并借助提示学习方法在预训练模型微调方面的优势,采用BERT和提示学习(Prompt)对藏文讽刺对话进行识别。实验结果表明,本文所提出的方法优于基于BERT的句子级讽刺数据集(无上下文信息)上的结果,并验证了该方法的可行性,望对未来的藏文讽刺识别任务提供有效借鉴。

In response to the difficulty of detecting ironic expressions in emotion classification research and the lack of research on Tibetan sarcasm recognition,a dataset of Tibetan sarcasm dialogue was created and the advantage of Prompt learning method in pre-training model fine-tuning was used to identify Tibetan sarcasm dialogue with BERT and Prompt.The experimental results show that the proposed method is superior to those on the BERT-based sentence-level satire dataset(no context information),and validate the feasibility of the proposed method,hoping to provide an effective reference for the Tibetan satire recognition task in the future.