Tensor Fusion Network for Multimodal Sentiment Analysis

摘要: Multimodal sentiment analysis is an increasingly popular research area, which extends the conventional language-based definition of to a multimodal setup where other relevant modalities accompany language. In this paper, we pose problem as modeling intra-modality and inter-modality dynamics. We introduce novel model, termed Tensor Fusion Network, learns both such dynamics end-to-end. The proposed approach tailored for volatile nature spoken language in online videos well accompanying gestures voice. experiments, our model outperforms state-of-the-art approaches unimodal analysis.