dataset. Compare with the result in Mo et al., 2020,
the
of GA-based method improves
because of fine-tuned batch size.
We conducted a paired t-test for accuracies
between each baseline and our proposed method. As
a result, we confirmed that our proposed method
outperforms the baselines, which is statistically
significant at p < 0.01. In addition, we confirmed that
our proposed method achieved the shortest execution
time. Notably, we do not compare the execution time
with the Random method because the method is not a
combinational optimization algorithm and has the
lowest recommendation accuracy.
We also experimented with different
to
confirm the effectiveness of shorting window size.
Because the DA completed the execution within 5
min, we set
to 5 min with the other time
parameters as in the previous setting (
= 6 h
and
= 4 h). As shown in Table 1, we confirmed
increased drastically as
was
shortened, which means that if the optimization
algorithm runs faster, the number of users that we
correctly predict their ad categories increase. Hence,
shortening the periodic optimization on DA is
important.
To summarize the experimental results, with
Logistic regression, we successfully shortened the
periodic advertisement recommendation from 525s to
108s and increased the accuracy from 0.239 to 0.324
compared to GA. With XGBoost, we also shortened
the execution time from 526s to 108s while
improving accuracy from 0.237 to 0.322.
5 CONCLUSION
In this paper, we proposed a new method, namely the
DA method, to optimize ads periodically in a short
period by using DA to solve the optimization
problem: maximizing CVR while satisfying the
delivery constraints, that is, the number of ads
delivered for each category. Our method consists of
two steps: 1) prediction to generate ad candidates for
each user, and 2) optimization of candidates to meet
the number of ad delivery constraints, which is
difficult to solve within an acceptable period on a
general-purpose computer. Experiments on a real
dataset showed that our proposed method
successfully improved the accuracy by shortening the
periodic advertisement recommendation: 0.239 to
0.324 with prediction algorithm Logistic regression
while shortening the execution time from 525s to
108s; and 0.237 to 0.322 with XGBoost while
shortening the execution time from 526s to 108s.
Our future plan includes conducting online tests
to verify the performance of our proposed model.
REFERENCES
Abrams, Z., Mendelevitch, O., and Tomlin, J., 2007.
Optimal delivery of sponsored search advertisements
subject to budget constraints. In Proceedings of the 8th
ACM conference on Electronic commerce, pp. 272-278.
Agarwal, D., Chen, B., and Elango, P., 2009. Spatio-
temporal models for estimating click-through rate. In
Proceedings of the 18th international conference on
World wide web, pp. 21-30.
Aramon, M., Rosenberg, G., Valiante, E., Miyazawa, T.,
Tamura, H., and Katzgrabeer, H., 2019. Physics-
inspired optimization for quadratic unconstrained
problems using a digital annealer. Frontiers in
Physics, 7(48), pp.1-14.
Chen, T., and Guestrin, C., 2016. XGBoost: A scalable tree
boosting system. In Proceedings of the 22nd ACM
SIGKDD International Conference on Knowledge
Discovery & Data Mining, pp. 785-794.
Goldberg, D. E., 1989. Genetic Algorithms in Search,
Optimization and Machine Learning. Addison-Wesley
Longman Publishing Company.
Grigas, P., Lobos, A., Wen, Z., and Lee, K., 2017. Profit
Maximization for Online Advertising Demand-Side
Platforms. in Proceedings of the ADKDD'17, pp. 1-7.
Huang, Z., Pan, Z., Liu, Q., Long, B., Ma, H., and Chen, E.,
2017. An Ad CTR Prediction Method Based on Feature
Learning of Deep and Shallow Layers. In Proceedings
of the 2017 ACM on Conference on Information and
Knowledge Management, pp. 2119-2122.
Juan, Y., Lefortier, D., and Chapelle, O., 2017. Field-aware
factorization machines in a real-world online
advertising system. In Proceedings of the 26th
International Conference on World Wide Web
Conference, pp. 680-688.
Kang, S., Jeong, C., and Chung, K., 2020. Advertisement
Recommendation System Based on User Preference in
Online Broadcasting. In Proceedings of 2020
International Conference on Information Networking,
pp. 702-706.
Mo, F., Jiao, H., Morisawa, S., Nakamura, M., Kimura, K.,
Fujisawa, H., Ohtsuka, M., and Yamana, H., 2020.
Real-Time Periodic Advertisement Recommendation
Optimization using Ising Machine. In Proceedings of
2020 IEEE International Conference on Big Data
(IEEE BigData 2020), 3pages (accepted as poster
presentation).
Pan, J., Xu, J., Ruiz, A., Zhao, W., Pan, S., Sun, Y., and Lu,
Q., 2018. Field-weighted factorization machines for
click-through rate prediction in display advertising. In
Proceedings of the 2018 World Wide Web Conference,
pp. 1349-1357.
Shan, L., Lin, L., and Sun, C., 2018. Combined Regression
and Tripletwise Learning for Conversion Rate
Prediction in Real-Time Bidding Advertising. in