Syntactic and Semantic Enhanced Text Generation Model for Aspect-based Sentiment Triplet Extraction
Xin Wei, Chengguo Lv · 2024
Aspect-based sentiment triplet extraction is a complex and practical task in sentiment analysis. The goal of this task is to extract aspect terms, opinion terms, and sentiment polarity from a text, such as restaurant reviews. These three sentiment elements form a sentiment triplet. Currently, the predominant methods for this task involve using span-level models, tagging schemes, or complex cascading networks. These methods have achieved good results, but the network structures are generally complex and diverse, which hinders the formation of a unified framework for ASTE. Additionally, there are still challenges in extracting sentiment triplets from complex sentences, such as overlapping sentiment elements and multiple sentiment triplets within a single sentence. For the above questions, we propose a text generation model which integrates syntax and semantics (MISS) to enhance the interaction between words in the text. Integrating syntax and part-of-speech information in dependency syntactic parsing and mining the semantic information contained in the model itself, not only enhances the interaction between words in the text but also simplifies the model architecture. we conducted experiments on multiple datasets, and the experimental results showed that our method is simple and effective, achieving performance comparable to state-of-the-art methods.