Learning interactions across sentiment and emotion with graph attention network and position encodings

Sentiment classification and emotion recognition are two close related tasks in NLP. However, most of the recent studies have treated them as two separate tasks, where the shared knowledge are neglected. In this paper, we propose a multi-task interactive graph attention network with position encodings, termed MIP-GAT, to improve the performance of each task by simultaneously leveraging similarities and differences. The main proposal is a multi-interactive graph interaction layer where a syntactic dependency connection, a cross-task connection and position encodings are constructed and incorporated into a unified graphical structure. Empirical evaluation on two benchmarking datasets, i.e., CMU-MOSEI and GoEmotions, shows the effectiveness of the proposed model over state-of-the-art baselines with the margin of 0.18%, 0.67% for sentiment analysis, 1.77%, 0.89% for emotion recognition. In addition, we also explore the superiority and limitations of the proposed model.