News is More than a Collection of Facts: Moral Frame Preserving News Summarization - model

We generate news summaries that preserve moral framing of the original article. In one of the tested approaches, we train a classifier to perform token-level classification of moral-laden words in news articles, and hand the predictions to a zero-shot summarization model with the instruction to preserve the detected words in the summary.

本研究生成保留原始新闻文章道德框架的新闻摘要。在其中一种测试方法中，我们训练分类器对新闻文章中的道德倾向词汇进行词元（Token）级分类，并将该分类器的预测结果输入至零样本（Zero-shot）摘要模型，同时为该模型设置需在生成的摘要中保留上述检测到的词汇的指令。