EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
4324. Bidding in Online Display Advertising: A Deep Reinforcement Learning Model for Mobile Gaming Market
Invited abstract in session MB-3: Optimization in Online Environments, stream Data Science Meets Optimization.
Monday, 10:30-12:00Room: 1005 (building: 202)
Authors (first author is the speaker)
1. | Qirui YANG
|
Department of Management Sciences, City University of Hong Kong | |
2. | Frank Chen
|
Management Sciences, The City Univ of Hong Kong | |
3. | Mengzhuo GUO
|
Sichuan University | |
4. | Houmin Yan
|
City University of Hong Kong | |
5. | Qingpeng ZHANG
|
The University of Hong Kong |
Abstract
When a user opens a website or app, there are often several advertisements interspersed. Generally, whenever an advertisement appears, loads, and is seen on a user's screen, it is referred to as an impression. For advertisers, an impression signifies an opportunity for their ads to be viewed. Therefore, real-time bidding (RTB) has emerged as a prominent paradigm, allowing advertisers to procure impressions through instantaneous automatic auctions. In RTB, the goal for advertisers is to maximize the total revenue generated from the impressions they win while considering constraints such as budget. Due to the difficulty in estimating the value of impressions and the bids of competitors, determining the optimal bid for each impression becomes a challenging problem for advertisers.
In our work, we first propose a novel definition of impression value under the background of the mobile gaming market, which takes into account players' in-game purchase, conversion revenue, and publisher characteristics. In addition, to enhance the interpretability of the strategy, we introduce a two-part method: the first part employs reinforcement learning techniques to obtain a two-dimensional vector based on an optimization problem, and the second part utilizes this vector to guide modifications to existing generic bidding paradigms in the industry. Real-world experiments demonstrate an increase in return on investment by over 80% compared to current practices after applying our method.
Keywords
- Auctions / Competitive Bidding
- Analytics and Data Science
- Machine Learning
Status: accepted
Back to the list of papers