Multimodal sentiment analysis based on multi-head attention mechanism
Xi Chen, Guanming Lu, Jingjie Yan
Abstract
Multimodal sentiment analysis is still a promising area of research, which has many issues needed to be addressed. Among them, extracting reasonable unimodal features and designing a robust multimodal sentiment analysis model is the most basic problem. This paper presents some novel ways of extracting sentiment features from visual, audio and text, furthermore use these features to verify the multimodal sentiment analysis model based on multi-head attention mechanism. The proposed model is evaluated on Multimodal Opinion Utterances Dataset (MOUD) corpus and CMU Multi-modal Opinion-level Sentiment Intensity (CMU-MOSI) corpus for multimodal sentiment analysis. Experimental results prove the effectiveness of the proposed approach. The accuracy of the MOUD and MOSI datasets is 90.43% and 82.71%, respectively. Compared to the state-of-the-art models, the improvement of the performance are approximately 2 and 0.4 points.