PSentScore: Evaluating Sentiment Polarity in Dialogue Summarization
This addresses the neglect of affective content in dialogue summarization, which is important for analyzing human interactions, but it is incremental as it focuses on evaluation and training adjustments.
The paper tackled the problem of evaluating how well dialogue summarization models preserve affective content, finding that state-of-the-art models perform poorly in this aspect, and showed that selecting training sets can improve affective preservation with a minor reduction in content metrics.
Automatic dialogue summarization is a well-established task with the goal of distilling the most crucial information from human conversations into concise textual summaries. However, most existing research has predominantly focused on summarizing factual information, neglecting the affective content, which can hold valuable insights for analyzing, monitoring, or facilitating human interactions. In this paper, we introduce and assess a set of measures PSentScore, aimed at quantifying the preservation of affective content in dialogue summaries. Our findings indicate that state-of-the-art summarization models do not preserve well the affective content within their summaries. Moreover, we demonstrate that a careful selection of the training set for dialogue samples can lead to improved preservation of affective content in the generated summaries, albeit with a minor reduction in content-related metrics.