Aspect-oriented Opinion Alignment Network for Aspect-Based Sentiment Classification

“Aspect-based sentiment classification is a crucial problem in fine-grained sentiment analysis, which aims to predict the sentiment polarity of the given aspect according to its context. Previous works have made remarkable progress in leveraging attention mechanism to extract opinion words for different aspects. However, a persistent challenge is the effective management of semantic mismatches, which stem from attention mechanisms that fall short in adequately aligning opinions words with their corresponding aspect in multi-aspect sentence

…

Taking this into consideration, some researchers have introduced various position information and proximity strategies to improve the effectiveness of aspect-based sentiment classification (ABSC) models. Gu et al. [10] proposed a position-aware bidirectional attention network (PBAN) that gives more attention to neighboring words of the aspect than words with long distances, while Zhou et al. [11] proposed a position-aware hierarchical transfer (PAHT) model that utilizes position information from multiple levels. Chen et al. [12] adopted a proximity strategy that assumes a closer opinion word is more likely to be the actual modifier of the target and designed a recurrent attention network (RAM) to counter irrelevant information using weight decay mechanisms. However, these approaches may not encompass or emphasize all relevant opinion words, limiting the model’s ability to fully comprehend the contextual meaning. Another trend of research has explored the use of graph neural networks (GNNs) for modeling syntactic structures of sentence based on dependency trees. For instance, Zhang et al. [23] introduced aspect specific graph convolutional networks (ASGCN) to handle aspect-level sentiment classification tasks. Tian et al. [24] proposed a type aware graph convolutional network (T-GCN) that utilizes an attentive layer ensemble to learn contextual information from different GCN layers. Li et al. [25] proposed a dual graph convolutional network (DualGCN) that simultaneously took syntax structures and semantic correlations into consideration. Although syntactic-based methods have achieved promising results, the imperfect parsing performance and randomness of input sentences inevitably introduce noise through the dependency tree.”