Emotion recognition of consumer comments based on graphic fusion
Lin Li, Minglan Yuan, Shangzhen Pang, Rongping Wang · International Journal of Information and Communication Technology · 2024
Multimodal sentiment classification is the use of image and text data to mine the potential emotions in consumer reviews. The picture scenes of the comments are complex, and there is a lot of information affecting the emotional judgement. To solve this problem, we propose a multimodal sentiment analysis model based on image translation to eliminate the redundant features of images and improve the efficiency of the model. Enhance text data to generate diverse data. Interactive learning realises the deep fusion of image and text. Leverage transformer to convert images into text by inputting them into space and output accurate consumer sentiment information in the images. Construct auxiliary sentences through translation, increase the amount of available text, and output classification results. Experiments were carried out on multi-ZOL, Twitter-2015 and 2017 datasets, and the accuracy rate reached 71.94%, 79% and 88.35%, respectively, which is far superior to other advanced methods.