運用資料探勘技術於廣告點擊率預測

Abstract

廣告是行銷產品的常見手法。根據統計，企業花費逾半的行銷費用於廣告行為。網際網路的興盛改變了廣告產業，廣告開始出現在數位媒體、網際網路等平臺。而網路廣告產業不斷發展之下，許多計費方式衍生而出。即時廣告競價是當代最熱門的廣告模式，能夠提升廣告商的利潤並為使用者進行個人化推薦。而點擊率是即時廣告競價中很重要的指標，決定使用者所看到的廣告內容。

本研究採用KDD Cup競賽資料進行點擊率預測實驗，試圖找出較佳的點擊率預測方法。實驗分別以數值預測演算法包含類神經網路、支持向量機、線性回歸等演算法及推薦方式進行預測模型建置，涵蓋資料前處理、屬性選擇、抽樣、離線運算及實際預測等，並比較各種預測模型之成效。最後發現數值預測演算法的效果優於協同過濾，其中最佳的是類神經網路。

Advertisement is a common approach in production marketing. According to statistics, enterprises spend more than half of its marketing expenses on advertising. Since the rise of Internet has changed the advertising industries, advertising began to appear in digital media, Internet, and other platforms. With the continuous growing of the advertising industries, many billing methods were derived. RTB (Real Time Bidding) is the most popular model which enhances not only profits for advertisers and publishers, but personalized recommendations for the users. In the meanwhile CTR (Click-Through Rate) also becomes an essential indicator which determines users what to see in RTB.

In this research, the designated experiment launches based on advertising recommendation history from KDD Cup competition. The goal of this work is to try to find the better CTR prediction approaches, using numeric prediction algorithms such as ANN (Artificial Neural Network), SVM (Support Vector Machine), linear regression and recommendation approach to complete experiment. All of the methods were compared and analyzed at the end of the experiment. Results showed that numeric prediction algorithm is better than collaborative filtering. For the three methods of numeric prediction, ANN has the best result.