2021/11/11 by Vikram Gupta, Gupta, Vikram · 1 citation
Computer Science · Psychology · #Computation and Language (cs.CL) #Emotion and Mood Recognition #FOS: Computer and information sciences #Sentiment Analysis and Opinion Mining #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2111.06181
openalex publication_date 2021/11/11 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28
Virtual Adversarial Training (VAT) has been effective in learning robust\nmodels under supervised and semi-supervised settings for both computer vision\nand NLP tasks. However, the efficacy of VAT for multilingual and multilabel\ntext classification has not been explored before. In this work, we explore VAT\nfor multilabel emotion recognition with a focus on leveraging unlabelled data\nfrom different languages to improve the model performance. We perform extensive\nsemi-supervised experiments on SemEval2018 multilabel and multilingual emotion\nrecognition dataset and show performance gains of 6.2% (Arabic), 3.8% (Spanish)\nand 1.8% (English) over supervised learning with same amount of labelled data\n(10% of training data). We also improve the existing state-of-the-art by 7%,\n4.5% and 1% (Jaccard Index) for Spanish, Arabic and English respectively and\nperform probing experiments for understanding the impact of different layers of\nthe contextual models.\n