chris90111/CriteoClickLogs

该数据集包含数百万展示广告的特征值和点击反馈，主要用于点击率（CTR）预测算法的基准测试。数据集由24个文件组成，每个文件对应一天的数据，覆盖了24天的Criteo流量。每行数据代表一个展示广告，第一列表示广告是否被点击（1为点击，0为未点击）。正例（点击）和负例（未点击）样本均经过子采样，但采样率不同以保护商业机密。特征包括13个整数特征（多为基于计数的数值属性）和26个类别特征（值被哈希为32位整数以实现匿名化，语义未公开）。数据按时间顺序排列，列以制表符分隔，缺失值字段留空。与Kaggle挑战数据集相比，该数据集在时间范围、子采样比率、特征顺序、特征计算方法和类别特征的哈希函数方面有所不同。This dataset contains feature values and click feedback for millions of display ads. Its primary purpose is to benchmark algorithms for clickthrough rate (CTR) prediction. It consists of 24 files, each corresponding to one day of data, spanning 24 days of Criteo traffic. Each row represents a display ad, with the first column indicating whether the ad was clicked (1) or not clicked (0). Both positive (clicked) and negative (non-clicked) examples have been subsampled at different rates to maintain business confidentiality. Features include 13 integer features (mostly count-based) and 26 categorical features (hashed into 32-bit integers for anonymization, with semantic meaning undisclosed). Rows are chronologically ordered, columns are tab-separated, and missing values are left empty. Compared to the Kaggle challenge dataset, it differs in time period, subsampling ratios, ordering of features, computation methods for some features, and hash function for categorical features.