Semi-IIN: Semi-supervised Intra-inter modal Interaction Learning Network for Multimodal Sentiment Analysis

Despite multimodal sentiment analysis being a fertile research ground that\nmerits further investigation, current approaches take up high annotation cost\nand suffer from label ambiguity, non-amicable to high-quality labeled data\nacquisition. Furthermore, choosing the right interactions is essential because\nthe significance of intra- or inter-modal interactions can differ among various\nsamples. To this end, we propose Semi-IIN, a Semi-supervised Intra-inter modal\nInteraction learning Network for multimodal sentiment analysis. Semi-IIN\nintegrates masked attention and gating mechanisms, enabling effective dynamic\nselection after independently capturing intra- and inter-modal interactive\ninformation. Combined with the self-training approach, Semi-IIN fully utilizes\nthe knowledge learned from unlabeled data. Experimental results on two public\ndatasets, MOSI and MOSEI, demonstrate the effectiveness of Semi-IIN,\nestablishing a new state-of-the-art on several metrics. Code is available at\nhttps://github.com/flow-ljh/Semi-IIN.\n