The content discusses the importance of understanding named entities in news captioning by leveraging commonsense knowledge. It introduces three modules: Filter Module, Distinguish Module, and Enrich Module, to enhance the understanding of named entities. The article highlights the challenges in distinguishing similar named entities and using external words to describe entities beyond the training data. It emphasizes the significance of incorporating commonsense knowledge for accurate entity understanding in news captioning.
Till ett annat språk
från källinnehåll
arxiv.org
Djupare frågor