Aspect-Based Sentiment Analysis Integrating Vector Interaction Mechanism

In recent years, considerable effort has been devoted to research on the relationship between aspect words and affective polarity. However, describing the relationship between aspect terms and phrase structure is challenging due to existing graph convolutional neural network models' over-reliance on syntactic parsing, which often neglects important semantic features. Based on research of aspect-based sentiment analysis to graph convolutional neural network models, we propose a sentiment analysis model that integrates a vector interaction mechanism and a cross-attention mechanism. The vector interaction layer is used to analyze dependency relationship, sentence structures and word weights as well as capturing syntactic relationship and enhancing classification accuracy in complex sentiment scenarios. Additionally, the cross-attention module describes the contextual relationship of aspect terms while feature extraction is performed using a Bidirectional Long Short-Term Memory network (BiLSTM). The fused features are input into the vector interaction layer for rich interactive feature extraction. Finally, average pooling is used to derive feature description, followed by a fully connected classifier for sentiment prediction. The Rest14 and Laptop14 datasets show that the designed model integrates syntactic, semantic, and component features effectively and accurately assesses sentiment polarity in complex expressions.