TRADE

<p>TRADE (<em>TRuly ADversarial ad understanding Evaluation</em>) is a new diagnostic test set consisting of 300 randomly selected image-based advertisements from the Pitt Ads dataset. Each ad is associated with 3 options:&nbsp;one valid explanation from the available annotations in the Pitt Ads dataset and two newly created adversarial negative explanations.&nbsp;<br><br>TRADE is presented in the following publication; please cite it if you use the dataset in your work:</p> <blockquote> <p>Anna Bavaresco, Alberto Testoni, and Raquel Fern&aacute;ndez, 2024. Don't Buy it! Reassessing the Ad Understanding Abilities of Contrastive Multimodal Models. In <em>Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL-2024)</em>. Association for Computational Linguistics.</p> </blockquote> <p>The paper can be found here: <a href="https://aclanthology.org/2024.acl-short.77/">https://aclanthology.org/2024.acl-short.77/</a></p>