An improved graphical convolutional network model based on interactive aspect-level sentiment analysis
Yuanhang Shi, Junjie Chen · 2023
Aspect-level text sentiment analysis is a more complete and specific task of text sentiment analysis, which determines the sentiment information expressed by a text with respect to each aspect. In recent years, the use of syntactic dependency trees as well as sentence structure information for aspect-level sentiment analysis has become more common, but they often ignore the role of sentence semantics involved in sentiment analysis and fail to combine them. In this paper, we improve and propose an interactive graph convolutional network (SInterGCN) model incorporating syntactic semantic structure for text-level sentiment analysis tasks. Specifically, a dependency graph is first constructed for each sentence in the syntactic relational dependency tree. This is followed by combining the graph of structural relations with the associated semantic graph, which is used to capture key aspect-oriented and contextual aspects. Furthermore, in order to interactively extract dependencies between aspect words and other aspects, an aspect-oriented GCN is used to model the representation learned by the aspect-oriented GCN based on the inter-aspect graph. Thus, the model can be aware of important contexts and aspect words when interactively learning aspect-specific sentiment features. Experimental results on four relevant benchmark datasets demonstrate that our proposed model outperforms the original interactive approach, achieving superior performance and validating the effectiveness of our model.