線上廣告即時競價之得標價預測

Winning Price Prediction in Real-Time Bidding for Online Advertising

劉幼菁(You-Jing Liu)

指導教授 ： 盧信銘

國立臺灣大學/管理學院/資訊管理學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201701505

摘要

預測線上廣告即時競價的得標價對於需求方平台而言，是參與競標前重要的工作項目，因為得標價相當於得標者最終所需支付的成本。本篇論文研究得標價的預測，使得需求方平台能夠藉此在即時競價市場中訂定適當的出價策略。然而，要精準地預測得標價相當困難，因為需求方平台無法觀察到過去未得標之競標的最終得標價。因此，本篇論文中提出一個混合模型，其結合分別由已知得標價的歷史已得標之競標資訊以及僅知出價的歷史未得標之競標資訊所訓練的兩個線性模型。此混合模型利用已知出價的歷史未得標之競標資料，來補足過去對於未得標之競標的得標價資訊空缺。最後，實驗結果顯示本篇論文所提出的混合模型對於得標價預測的結果優於線性迴歸模型及隨機生產前緣模型。

關鍵字

線上廣告 ； 即時競價 ； 需求方平台 ； 得標價預測 ； 展示型廣告

並列摘要

From the viewpoint of a Demand-Side Platform (DSP), forecasting the winning price is an important task before bidding an ad impression because the winning price is equivalent to the cost that a DSP must pay after winning a bid. This paper studies on how to predict the winning price for an ad impression so that a DSP can win the ad impression by offering a suitable bidding price in the Real-Time Bidding (RTB). However, it is difficult to accurately estimate winning price for a DSP because the winning price is unobserved when a DSP lost the bid. Therefore, we propose a mixture model that is composed of two regression models learning from winning bids with observable winning price and losing bids with observable bidding price, respectively. The mixture model takes advantage of observable bidding price of historical losing bids to reconstruct the missing distribution of winning price. Last, the results of experiments show that the proposed mixture model outperforms linear regression model and stochastic production frontier in terms of winning price prediction.

並列關鍵字

Online Advertising ； Real-Time Bidding ； Demand-Side Platform ； Winning Price Prediction ； Display Advertising

參考文獻

Aigner, D., Lovell, C. K., & Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, 6(1), 21-37.

Balakrishnan, R., & Bhatt, R. P. (2015). Real-time bid optimization for group-buying ads. ACM Transactions on Intelligent Systems and Technology (TIST), 5(4), 62.

Edelman, B. (2010). The design of online advertising markets. Handbook of Market Design.

Greene, W. H. (2008). The econometric approach to efficiency analysis. The measurement of productive efficiency and productivity growth, 1, 92-250.

Kumbhakar, S. C., & Lovell, C. K. (2003). Stochastic frontier analysis. Cambridge University Press.