A multichannel fusion learning model with syntax for Chinese-oriented aspect-level sentiment classification

Aspect-level sentiment classification (ASC), as a classic fine-grained sentiment analysis task, aims to analyze the sentiment polarity of different aspect terms in a sentence or document, which is a subtask of Aspect-based sentiment analysis (ABSA). Although many models of ASC have been proposed, they fail to effectively identify the range of local context and cannot fully and effectively utilize the semantic information of the global context. Moreover, there are few studies on Chinese ASC tasks. In this work, a multichannel fusion learning model with syntax for Chinese-oriented aspect-level sentiment classification is proposed. It uses the syntactic relative distance (SRD) computed based on a dependency syntax parsing tree to represent semantic distance. In addition, the model effectively integrates the semantic information of local context channel, global context channel and dependency tree channel, thereby improving the performance of the model. Compared with existing models, the proposed model can identify the range of local context more accurately and better explain the syntactic correlation between aspect terms and local context. Furthermore, the model can accurately analyze Chinese and English reviews. Experiment results on six English benchmark datasets and three Chinese benchmark datasets show that our model outperforms the baseline model and has better interpretability.