Authors:
Fan Mo
1
;
Huida Jiao
1
;
Shun Morisawa
1
;
Makoto Nakamura
2
;
Koichi Kimura
2
;
Hisanori Fujisawa
2
;
Masafumi Ohtsuka
3
and
Hayato Yamana
4
Affiliations:
1
Dept. of Computer Science and Communications Engineering, Waseda University, Tokyo, Japan
;
2
Fujitsu Laboratories Ltd., Kanagawa, Japan
;
3
Geniee, Inc., Tokyo, Japan
;
4
School of Science and Engineering, Waseda University, Tokyo, Japan
Keyword(s):
Computational Advertisement, Advertisement Recommendation, Digital Annealer, Real-time Bidding.
Abstract:
For commercial companies, tuning advertisement delivery to achieve a high conversion rate (CVR) is crucial for improving advertising effectiveness. Because advertisers use demand-side platforms (DSP) to deliver a certain number of ads within a fixed period, it is challenging for DSP to maximize CVR while satisfying delivery constraints such as the number of delivered ads in each category. Although previous research aimed to optimize the combinational problem under various constraints, its periodic updates remained an open question because of its time complexity. Our work is the first attempt to adopt digital annealers (DAs), which are quantum-inspired computers manufactured by Fujitsu Ltd., to achieve real-time periodic ad optimization. With periodic optimization in a short time, we have much chance to increase ad recommendation precision. First, we exploit each user’s behavior according to his visited web pages and then predict his CVR for each ad category. Second, we transform the
optimization problem into a quadratic unconstrained binary optimization model applying to the DA. The experimental evaluations on real log data show that our proposed method improves accuracy score from 0.237 to 0.322 while shortening the periodic advertisement recommendation from 526s to 108s (4.9 times speed-up) in comparison with traditional algorithms.
(More)