Multivariate graph neural networks on enhancing syntactic and semantic for aspect-based sentiment analysis

Aspect-based sentiment analysis (ABSA) aims to predict sentiment orientations towards textual aspects by extracting insights from user comments. While pretrained large language models (LLMs) demonstrate proficiency in sentiment analysis, incorporating syntactic and semantic features into ABSA remains a challenge. Additionally, employing LLMs for sentiment analysis often requires significant computational resources, rendering them impractical for use by individuals or small-scale entities. To address this, we propose the semiotic signal integration network (SSIN), which effectively combines syntactic and semantic features. The core syncretic information network leverages isomorphism and syntax to enhance knowledge acquisition. The semantically guided syntactic attention module further enables integrated semiotic representations via sophisticated attention mechanisms. Experiments on the publicly available SemEval dataset show that SSIN performs better than existing state-of-the-art ABSA baselines and LLMs such as Llama and Alpaca with high accuracy and macro-F1 scores. Moreover, our model demonstrates exceptional interpretability and the ability to discern both positive and negative sentiments, which is vitally important for real-world applications such as social media monitoring, health care, and customer service. Code is available at https://github.com/AmbitYuki/SSIN.