Automatic Creative Selection With Cross Modal Matching


Application developers advertise their Apps by creating product pages with App images, and bidding on search termsApplication developers advertise their Apps by creating product pages with App images, and bidding on search terms.
Apple Machine Learning Research 12:05 am on May 23, 2024


Automatic Creative Selection with Cross-Modal Matching explores the use of LXMERT pre-trained models fine-tuned for matching App images to search terms, outperforming other approaches. The research demonstrates significant improvements in advertiser and human relevance scores using this novel approach compared to baselines like Transformer+ResNet models.

  • LXMERT pre-trained model fine-tuned for cross-modal matching
  • Outperforms CLIP, Transformer+ResNet baseline by significant margins in both advertiser and human relevance scores
  • Two datasets used: advertisers' associated labels, and human ratings for image-text pairs
  • Applicability to various machine learning tasks through refined synthetic images using the presented method
  • Published in May 2024 with authors from Apple Research Machine Learning division

https://machinelearning.apple.com/research/automatic-creative-selection

< Previous Story     -     Next Story >

Copy and Copyright Pubcon Inc.
1996-2024 all rights reserved. Privacy Policy.
All trademarks and copyrights held by respective owners.