B-012
PNASNet-5およびWide&Deep Mixアプローチを用いたバナー広告のCTR分類

○平岩篤信・渡邊太郎・ディン・マイン グエン・チョン・タン ディン・ディン・タオ グエン・神谷　寛・大西一貫・マハモドゥル ハサン（アイレップ）

Click Through Rate (called CTR) based banner classification using PNASNet-5 with wide and deep mix modeling approach is proposed for improvement of prediction accuracy being applied to real display network internet banner ads considering advertisement strategies. The proposal provides prediction for using evaluation quality of banner image before actual broad casting and enhances realization to optimized broadcasting allocation for internet business advertisement. Accuracies of 87.3% and 95.9% are confirmed in two media platform cases comparing test between predicted CTR score and actual CTR score in performance report. The proposal helps image creative process via reducing revision and improves business speed.