Adform click prediction dataset

This data is a sample of Adform's ad traffic. Each record corresponds to an ad impression served by Adform, and consists of a single binary label (clicked/not-clicked) and a selected subset of features (c0-c9). The positives and negatives are downsampled at different rates. The data is chronologically ordered. The file is gzipped and each line corresponds to a single record, serialized as JSON. The JSON has the following fields: \"l\": The binary label indicating whether the ad was clicked (1) or not (0). \"c0\" - \"c9\": Categorical features which were hashed into a 32-bit integer. The semantics of the features are not disclosed. The values are stored in an array, because some of the features have multiple values per record. When a key is missing, the field is empty. The files are named \"adform.click.2017.xx.json.gz\", where \"xx\" is the index (01-05). The files are indexed chronologically, and the records (lines) in the file within are ordered chronologically.

本数据集为Adform广告流量样本集。每条记录对应一次Adform投放的广告展示（ad impression），包含一个二元标签（点击/未点击）以及经筛选的特征子集（c0至c9）。正负样本以不同比例进行了下采样。数据集按时间顺序排列。该数据集文件采用gzip压缩（gzip）格式，每行对应一条以JSON格式序列化的记录。该JSON包含以下字段："l"：用于标识广告是否被点击的二元标签（1代表已点击，0代表未点击）；"c0"至"c9"：已被哈希为32位整数的分类特征，特征的具体语义未公开。由于部分单条记录可能包含多个特征值，因此特征值以数组形式存储。若某键缺失，则对应字段为空。数据集文件命名格式为"adform.click.2017.xx.json.gz"，其中"xx"为文件索引（01至05）。所有文件按时间顺序编号，文件内的记录（行）同样按时间顺序排列。