Contextual multimodal sentiment analysis with information enhancement
Dandan Jiang, Dongsheng Zou, Zhipeng Deng, Jingpei Dan · Journal of Physics Conference Series · 2020
Abstract The results of sentiment analysis research can be applied to various social situations, including human communication understanding and dialogue system. Multimodal sentiment analysis involves the identification of sentiment in videos. This paper proposes a multimodal sentiment analysis framework, in which the Attention-based Bi-directional Gated Recurrent Unit (AT-BiGRU) model is used to obtain the contextual relationship among utterances, and the Information Enhancement Fusion (IE-Fusion) model is used to fuse the multimodal features of each utterance. Experimental results show that our framework achieves 82.85% and 65.77% accuracies on two different benchmark datasets, CMU-MOSI and IEMOCAP, respectively. Our method outperforms the state-of-the-art models on the datasets.