融合语言特征的多模态中文反讽识别模型

打开文本图片集

中图分类号：TP391 文献标志码：ADOI：10. 13705/j. issn. 1671-6841.2024096

文章编号：1671-6841（2025）05-0016-08

Abstract：In light of the disparity between graphic and text modes and insufficient atention to textual information，a multimodal Chinese sarcasm detection model integrated with linguistic features was proposed.The Chi-square statistical method was used to extract words with sarcastic and non-sarcastic meanings，forming the linguistic feature system. TextCNN was utilized to extract linguistic features，enhancing the distinction between sarcastic and non-sarcastic characteristics. TextCNN and ResNet were employed to extract text and image features，and a cross-attention mechanism was introduced. Residual connections were used to fuse text and image features，to help preserve language characteristics.The effectiveness of the proposed model was verified by using an emergency multimodal dataset containing sarcastic comments.The results showed that the model outperformed the baseline model，and focusing on textual linguistic features helped improve the efficiency of problem-solving.

Key Words： linguistic feature; Chinese sarcasm detection； emergency event; multimodality

0 引言

情趋势，甚至导致网络舆情发展不可控。（剩余13213字）