智能计算机与应用
智能計算機與應用
지능계산궤여응용
Intelligent Computer and Applications
2015年
5期
22-25,28
,共5页
卷积神经网络%点击率预测%搜索广告
捲積神經網絡%點擊率預測%搜索廣告
권적신경망락%점격솔예측%수색엄고
Convolution Neural Network%Click-Through Rate Prediction%Search Advertising
广告点击率的预测是搜索广告进行投放的基础。目前已有的工作大多数使用线性模型或基于推荐方法的模型解决点击率预测问题,但这些方法没有对特征之间的关系进行深入的探索,无法完全体现广告点击预测中各个特征之间的关系。本文提出了基于卷积神经网络的搜索广告点击率预测的方法,阐述了卷积神经网络在特征的学习上模拟人的思维过程,并进一步分析了不同特征在广告点击率预测中的作用,在KDD Cup 2012中Track 2数据集上的实验结果验证了本文提出的方法能够提高搜索广告点击率的预测效果,其AUC值达到0.7925。
廣告點擊率的預測是搜索廣告進行投放的基礎。目前已有的工作大多數使用線性模型或基于推薦方法的模型解決點擊率預測問題,但這些方法沒有對特徵之間的關繫進行深入的探索,無法完全體現廣告點擊預測中各箇特徵之間的關繫。本文提齣瞭基于捲積神經網絡的搜索廣告點擊率預測的方法,闡述瞭捲積神經網絡在特徵的學習上模擬人的思維過程,併進一步分析瞭不同特徵在廣告點擊率預測中的作用,在KDD Cup 2012中Track 2數據集上的實驗結果驗證瞭本文提齣的方法能夠提高搜索廣告點擊率的預測效果,其AUC值達到0.7925。
엄고점격솔적예측시수색엄고진행투방적기출。목전이유적공작대다수사용선성모형혹기우추천방법적모형해결점격솔예측문제,단저사방법몰유대특정지간적관계진행심입적탐색,무법완전체현엄고점격예측중각개특정지간적관계。본문제출료기우권적신경망락적수색엄고점격솔예측적방법,천술료권적신경망락재특정적학습상모의인적사유과정,병진일보분석료불동특정재엄고점격솔예측중적작용,재KDD Cup 2012중Track 2수거집상적실험결과험증료본문제출적방법능구제고수색엄고점격솔적예측효과,기AUC치체도0.7925。
Click-Through Rate ( CTR) prediction is the foundation of search advertising. Nowadays, lots of researches have been explored to predict CTR, and most of those researches either rely on liner model or employ method of recommen-dation system. However, the relations between different features in CTR predication have not been fully explored in previ-ous works, and the relations between different features also cannot be fully embodied. In this paper, CTR prediction for search advertising based on convolution neural network is proposed, and process of convolution neural network simulating the process of human thought on feature learning is explained. Furthermore, the performance of different features have been analyzed in the task of predicting CTR. Experiments are conducted on the dataset of KDD Cup 2012 Track2 and the pro-posed method achieves 0. 792 5 in AUC, demonstrating the effectiveness of the proposed approach.