MGTS: An Efficient Method for Constructing Multilingual Entity-Opinion Sentiment Triples

Entity Sentiment Triple Extraction (ESTE) aims to automatically extract sentiment triples related to specific entities from text. The internet contains an immense amount of multilingual opinion information. Due to the complexity of different languages and the diversity of opinion expressions, conducting multilingual ESTE tasks presents a significant challenge for existing models. To address this issue, we propose an efficient ESTE framework. This method constructs a grid representation suitable for multilingual tasks by concatenating word pair embeddings and enhances the grid tagging decoding algorithm to solve the ESTE problem in an end-to-end manner. Experimental results demonstrate that, compared to several related classical and state-of-the-art triple extraction methods, our approach performs well in both opinion extraction accuracy and multilingual sentiment analysis. This validates the efficacy of our method in multilingual contexts.