基于多模态深度学习的旅游评论反讽识别研究

Research on Irony Recognition of Travel Reviews Based on Multi-modal Deep Learning

[目的/意义]反讽作为一种隐性而间接的情感表达方式,在社交评论中被广泛使用,正确识别用户的反讽情感对于网络平台和服务商具有重要意义。[方法/过程]文章基于深度学习构建多模态反讽识别模型,以在线旅游评论为例,运用深度学习模型分别提取文本、表情符号和图片的特征向量,通过不同的特征融合方式进行反讽识别。[结果/结论]通过与单模态反讽识别模型进行对比实验,发现文章提出的多模态旅游评论反讽识别模型在准确率、召回率等指标上的结果更优,验证了多模态模型比单模态模型反讽识别效果更佳的结论。

[Purpose/significance]As an implicit and indirect emotional expression,irony is widely used in social comments.Correctly identifying users’irony emotion is of great significance to network platforms and service providers.[Method/process]In this paper,a multi-modal irony recognition model is constructed based on deep learning.Taking online travel reviews as an example,the feature vectors of texts,emojis and pictures are extracted respectively by deep learning model,and irony recognition is carried out through different feature fusion methods.[Result/conclusion]Through comparative experiments with single-modal irony recognition model,it is found that the multi-modal travel review irony recognition model proposed in this paper has better results in terms of Accuracy,Recall and other evaluation indicators,which verifies that the multi-modal model is better than the single-modal model at irony recognition.