alimamaTech/TRACE

---
license: apache-2.0
task_categories:
- tabular-classification
size_categories:
- 100M<n<1B
---
We introduce TRACE, the first benchmark dataset for post-click GMV prediction with delayed feedback.

📁 Data structure includes:
1. **Features**:
* **User/Item/Contextual Attributes**: 22 such attributes are provided (e.g., `feature_0` to `feature_21`).
* **Click Timestamp**: `click_time`.

2. **Labels and Sequence Information**:
* **Gross Merchandise Volume (GMV) Sequence**: The sequence of transaction amounts for all purchases associated with a click. This is represented by `dirpay_amt` for the current purchase and `prev_dirpay_amt` for previous purchases.
* **Purchase Count Sequence**: The sequence indicating the order of each purchase within the attribution window for a given click. `count` represents the current purchase's order, and `total_counts` indicates the final number of purchases.
* **Purchase Timestamp Sequence**: The sequence of timestamps for each purchase. `pay_time` typically represents the timestamp of the current purchase, and `prev_pay_time` contains timestamps of previous purchases.
* **Repurchase Indicator (`multi_tag`)**: A binary label (0 or 1) indicating whether the GMV generated by a click is from a single purchase (0) or a repurchase (1). This is derived from the number of purchases.
* **Final Ground-Truth GMV Label**: The total GMV accumulated by the end of the attribution window. This is calculated as the sum of `dirpay_amt`.

许可证：Apache-2.0
任务类别：表格分类（tabular-classification）
样本规模：100M<n<1B

我们推出了TRACE，这是首个针对带延迟反馈的点击后商品交易总额（Gross Merchandise Volume，GMV）预测的基准数据集。

📁 数据结构包含：
1. **特征**：
* **用户/商品/上下文属性**：共包含22类此类属性（例如`feature_0`至`feature_21`）。
* **点击时间戳**：字段名为`click_time`。

2. **标签与序列信息**：
* **GMV序列**：指与某次点击相关的所有购买交易金额序列，其中当前交易金额由`dirpay_amt`表示，过往交易金额由`prev_dirpay_amt`表示。
* **购买次数序列**：用于指示单次点击归因窗口内各笔购买的下单顺序。`count`代表当前购买的顺位，`total_counts`则表示最终总购买笔数。
* **购买时间戳序列**：各笔购买的时间戳序列，`pay_time`通常指代当前购买的时间戳，`prev_pay_time`存储过往购买的时间戳。
* **复购标识（`multi_tag`）**：二分类标签（0或1），用于标识某次点击产生的GMV来自单笔购买（0）还是复购（1），该标签由总购买笔数推导而来。
* **最终真实GMV标签**：归因窗口结束时累计的总GMV，计算方式为所有`dirpay_amt`的求和结果。