Multimodal Entity Linking Evaluation Dataset for Art (Version 3.0)

MELArt Dataset. The dataset adds named entity linking annotations to the sentences in the Artpedia dataset (https://aimagelab.ing.unimore.it/imagelab/page.asp?IdPage=35). The files inside MELArt contain the following information: el_candidates.jsonl: all the candidates, each line is a json file containing the basic information extracted from Wikidata for each candidate. melart_annotations.json: contains the full set of annotations. Each element is a painting that includes the basic information from Artpedia, the depictions extracted from Wikidata, and the annotated mentions for each of the sentences. Each painting has a corresponding split and the annotations from the test split are manual annotations. melart_automatic_annotations.json: contains the automatically generated annotations before integrating the manual annotations. images/image_urls.txt: Each line corresponds to the name of the file for Wikimedia Commons or the full URL of images not part of Commons needed for the dataset. For downloading the images we recommend to use the image crawler from the Github repository: https://github.com/HPI-Information-Systems/MELArt/blob/main/crawl_images.py The full code used to produce the dataset can be found at https://github.com/HPI-Information-Systems/MELArt