DiaASQ : A Benchmark of Conversational Aspect-based Sentiment Quadruple Analysis

In this paper, we consider filling the gap of ple are scattered around the whole conversation dialogue-level ABSA. We follow the line of recent due to the complex replying structure, which requires quadruple ABSA and present a task of conversational the model to do cross-utterance extraction. DiaASQ sets the goal to detect DiaASQ framework. The task aims to extract the DiaASQ data indicate that our model shows significant three quadruples over the dialog: ('Xiaomi 11', superiority than several strong baselines. 'WiFi module', 'bad design', 'negative'), ('Xiaomi To sum up, this work contributes in threefold: 11', 'battery life', 'not well', 'negative') and ('Xiaomi We pioneer the research of dialogue-level 6', 'screen quality', 'very nice', 'positive'). Specifically, To benchmark the task, we manually annotate a we introduce a conversational aspect-based large-scale DiaASQ dataset.