基於局部性預測即時競價之得標價

Abstract

現今的廣告業模式已和以往大不相同，即時競價的數位廣告技術，簡單來說就是一個平台，廣告主可以在任何時間，對任何人所開放的網頁廣告競標以播放廣告的權利，線上廣告跟傳統廣告相比更能掌握到使用者族群，因此成為目前最受歡迎的廣告平台之一。 即時競價系統，大致上包含兩種角色：廣告主（Advertiser）和媒體（Publisher）。媒體能提供廣告版位，讓廣告主們即時競標，贏得競標後會有一個得標價，得標價與廣告主是否可得到該版位以及廣告主所需花的成本息息相關，因此預測得標價是即時競價中一項重要的研究課題，也是本篇論文的研究方向。傳統上使用線性迴歸法來預測得標價，但忽略了大量的設限資料，我們基於資料集局部性的特性，使用局部加權迴歸散點平滑法，其中資料集包含設限資料，我們也利用標籤傳播法和Locality Sensitive Hashing法估算設限資料的價格，實驗結果顯示我們的預測模型比現有的方法來的準確。實驗中除了比較其他方法，後續也探討了不同的競標其間的差異，以及不同的產業類別之間其預測的結果也不盡相同。

In recent years, digital advertisement technique has become more and more popular due to the growth of the Internet. Real-time bidding（RTB）, a kind of advertisement type, has been widely used in online advertisement. RTB is an advertising inventory, which is bought and sold on a per-impression basis, via programmatic instantaneous auction, similar to financial. Demand-Side Platform（DSP）is a part of RTB, which gives buyers to access multiple sources of inventory, and winning price prediction is essential to DSP, since it determines whether the DSP can win the bid by placing a proper bidding value in the real-time bidding auction. A major challenge is that DSP usually suffers from the censoring of the winning price, especially for those lost bids in the past. In this thesis, we propose to use label propagation and locality sensitive hashing to predict the winning price of censored data. Additionally, we propose to apply locally weight scatterplot smoothing method to predict the winning price owing to the locality characteristic of the dataset. The experimental results show that the proposed method, which consider the censored data and locality, outperforms the alternatives in terms of the prediction accuracy.