Hybrid sentiment analysis with textual and interactive information

In the past decade, various methods have been proposed to address document-level sentiment classification. However, the exploration of user–product interactions has not been sufficiently studied in the literature. In this work, we aim to investigate the effectiveness of exploiting user–product relations, and propose a hybrid semantic and interactive model for the classification task. The novelty of the proposed method is a ranking graph module and a latent matching module, where the former is capable of capturing high-order connectivity among the nodes, while the later is able to preserve semantics of local connectivity during the recursive graph learning processing. These two modules are seamlessly incorporated, enabling the proposed model to learn comprehensive and discriminative representations for the specific classification task. We conduct extensive experiments on three public datasets, and demonstrate the advantage of the proposed model over the state-of-the-art baselines.