[2M1-GS-10] AI application:

[2M1-GS-10-05] Prediction of advertisement delivery effectiveness in large action models using creative information

〇Manami Kawasaki¹, Eri Nakahara², Hisashi Kurasawa², Nobuhiro Oki², Yuki Taniguchi², Rin Kan², Takeshi Kato¹, Daisuke Ochi¹ (1. NTT DOCOMO, INC., 2. Nippon Telegraph and Telephone Corporation)

Keywords:Transformer, Customer Behavior Prediction, Advertisement

In recent years, companies have been working to optimize ad creatives to increase the effectiveness of ad delivery, selecting and delivering creatives with high delivery effectiveness. However, when new ad creatives are used for advertisement delivery, it is difficult to predict the effectiveness of delivery because there are no past delivery results.
This paper proposes a method for indirectly predicting the effectiveness of new advertisement delivery by using ad creatives used in other past campaigns and their delivery results to address this issue. We verified whether it is possible to indirectly predict the effectiveness of advertisement delivery for new ad creatives by learning the results of advertisement delivery for other past measures. The validation results show that the proposed method can predict the effectiveness of new advertisement delivery, and that the weights of the prediction scores and creative information contribute to the accuracy of the prediction.