CGSL: Collaborative Graph and Segment Learning Based Aspect-Level Sentiment Analysis Model

Aspect-level sentiment analysis identifies emotional polarity from the context of specific aspect words. Most of the current research on aspect-level sentiment analysis focuses on mining the grammatical and semantic relationship between aspects and isolated sentences. However, the relationship between words and multiple sentence contexts in the whole corpus and the sentiment attributes of different segments are ignored. We propose a collaborative graph and segment learning based model (CGSL) to explore the relationship between aspects and multiple sentence context to tackle these problems. First, we propose a collaborative graph interaction component that applies a gate to combine the collaborative graph with the graph convolution network to control the information transfer. Second, to process the emotional attributes of different segments, we propose segment learning component in the model, simulation agent judgment, and strategy gradient method optimization to improve the performance. Finally, the output of the collaborative graph interaction component and segment learning component is integrated with the output of the attention mechanism as the final output of the proposed model. The experimental results on multiple datasets show that our model outperforms the state-of-the-art non-pre-trained models.