Conversational Aspect-Based Sentiment Quadruple Analysis with Consecutive Multi-view Interaction

With the development of information technology, the increasing amount of content on the web has made aspect-based sentiment analysis an essential tool for extracting information about emotional states. However, most of the existing work focuses on a single text, while little attention is paid to the task of sentiment analysis in complex texts such as dialogues, in which the quadruple of target-aspect-opinion-sentiment may appear in different speakers during one conversation thread. In this paper, we proposed a novel framework that was built by a Chinese pre-trained language model and a grid tagging classifier. In addition, we use multi-view interaction with three consecutive multi-head attention modules to improve the performance and robustness of our model. Besides, based on the excellent performance of the Chinese pre-training model, the English version is transferred from the final Chinese weights to achieve cross-lingual transfer. To improve the generalization ability of the model, cross-validation is used to select the best one. Our model ranks first on track 4 of the NLPCC-2023 shared task on conversational aspect-based sentiment quadruple analysis. Our code is publicly available at https://github.com/Joint-Laboratory-of-HUST-and-PAIC/nlpcc2023-shared-task-diaASQ.