Good News, Everyone! Context driven entity-aware captioning for news images
This addresses the limitation of current image captioning systems for news applications by producing captions that incorporate contextual knowledge, though it is incremental in leveraging existing captioning methods with added context.
The authors tackled the problem of generating more interpretative captions for news images by integrating contextual information from associated articles, achieving state-of-the-art results on their newly introduced 'GoodNews' dataset.
Current image captioning systems perform at a merely descriptive level, essentially enumerating the objects in the scene and their relations. Humans, on the contrary, interpret images by integrating several sources of prior knowledge of the world. In this work, we aim to take a step closer to producing captions that offer a plausible interpretation of the scene, by integrating such contextual information into the captioning pipeline. For this we focus on the captioning of images used to illustrate news articles. We propose a novel captioning method that is able to leverage contextual information provided by the text of news articles associated with an image. Our model is able to selectively draw information from the article guided by visual cues, and to dynamically extend the output dictionary to out-of-vocabulary named entities that appear in the context source. Furthermore we introduce `GoodNews', the largest news image captioning dataset in the literature and demonstrate state-of-the-art results.