<p>The Dialogue-level Aspect-based Sentiment Quadruple Analysis (DiaASQ) aims to extract sentiment and opinions from multi-turn and multi-party dialogues. This is crucial for improving products and services. However, existing models face challenges in resisting noise introduced by colloquial expressions, adequately capturing the complex contextual semantics, and effectively distinguishing the semantics of different views. This work proposes a multi-view adaptive dynamic interaction model (MADIAT) integrating adversarial training for the DiaASQ task. Firstly, the adaptive adversarial training module adopts various adversarial training strategies tailored to the characteristics of different datasets to enhance the model’s robustness against noise in dialogue texts. Secondly, for the uncertainties and variations in dialogue length, sentiment shift, and topic distribution, the multi-view adaptive dynamic interaction module effectively capture the complex fine-grained semantics of different views by dynamically adjusting the attention and activation values of semantics from each view. Finally, a multi-view semantic fusion module assigns different weights to each view to highlight their importance, facilitating the decoding module in better accomplishing sentiment quadruple analysis. Experimental results on both the Chinese (ZH) and English (EN) datasets demonstrate that the proposed model achieves competitive overall performance for the DiaASQ task, validating its effectiveness in dialogue-level sentiment analysis. Notably, the model outperforms existing approaches across all three subtasks on the Chinese dataset.</p>