ALSEM: aspect-level sentiment analysis with semantic and emotional modeling

Aspect-based sentiment analysis is a fine-grained sentiment analysis task that involves classifying the sentiment polarity of specific aspect terms in sentences. Existing semantic graph convolutional networks fail to accurately capture the relationship between aspect terms and opinion words, thus overlooking attention to aspect terms and leading to inaccurate classification results. To address this issue, this paper proposes Aspect-Level Sentiment Analysis with Semantic and Emotional Modeling (ALSEM). The model theoretically establishes a framework that systematically explains the interaction between semantic information and sentiment knowledge, thereby guiding the design and method selection for the model. By integrating self-attention and aspect-aware attention mechanisms, ALSEM constructs an attention score matrix for sentences and uses graph convolutional networks to extract semantic features based on this matrix. The model captures both aspect-related semantic associations and global semantic information, providing comprehensive support for sentiment classification. Additionally, it incorporates external sentiment knowledge to enhance the interaction between aspect terms and opinion words. We conduct experiments on three benchmark datasets to evaluate the performance of the proposed model. Experimental results demonstrate that on the Restaurant14, Laptop14, and Twitter datasets, our proposed model achieves accuracy rates of 85.26%, 80.32%, and 76.72%, respectively.