README.md

---
license: cc-by-4.0
task_categories:
- tabular-classification
- reinforcement-learning
language:
- en
- multilingual
tags:
- recommendation-system
- recommendation
- machine-learning
- email
- tabular
- marketing
- click-through-rate-prediction
pretty_name: XCampaign Dataset
size_categories:
- 10M<n<100M
---

# XCampaign Dataset
<h4>
	<a href="https://github.com/zidcenek/Active-Learning-for-Email-Interaction-Dynamics" target="_blank"> 💻Github Repo</a>
	<a href="https://dl.acm.org/doi/10.1145/3746252.3760832" target="_blank">📖Paper Link</a>
</h4>

## Introduction
This repository contains the Mailprofiler's **XCampaign Dataset** -- provided by Mailprofiler; 
[XCampaign](https://xcampaign.info/switzerland-en/) represents an email campaign management platform. 
The dataset was published alongside our CIKM 2025 paper *Active Recommendation for Email Outreach Dynamics*. 

The dataset of almost 15 million interactions captures user-level interactions with periodic marketing mailshots, 
including whether an email was opened and the time-to-open (TTO).

## Dataset and Fields
The **XCampaign Dataset** includes the following fields:
- `mailshot_id`: (or template id) identifier of the mailshot campaign
- `user_id`: anonymized recipient identifier
- `opened`: binary label (\(1\) if opened, \(0\) otherwise)
- `time_to_open`: time delta between send and open (a parseable string of a timedelta `0 days 09:39:32`)

## Global Statistics
All statistics below are computed from the full dataset.
- `Rows`: 14,908,085; `Users`: 131,918; `Mailshots`: 160
- Global open rate: 9.09%
- Per-mailshot open rate: $9.13\% \pm 3.58\%$
- Per-user open rate: mean $12.33\% \pm 20.46\%$
- Time-to-open (opened only): mean 1d 17h 25m; median 6h 25m
- Fraction opened within 1h: 25.9%; within 24h: 71.2%; within 7d: 93.0%
- Sent to users at each mailshot: $93,175 \pm 19,162$
- Item \(\times\) User interaction matrix density: 70.63%


## How to Use and Cite

The XCampaign Dataset is made available under the **Creative Commons Attribution 4.0 International License (CC BY 4.0)**.

This license allows you to share and adapt the dataset for any purpose, **including commercial use**, as long as you provide appropriate credit.

If you use this dataset in your work, please **cite the following paper**, which introduced the dataset:

### Plain Text Citation
> Čeněk Žid, Rodrigo Alves, and Pavel Kordík. 2025. Active Recommendation for Email Outreach Dynamics. In *Proceedings 
> of the 34th ACM International Conference on Information and Knowledge Management (CIKM '25)}*. Association for 
> Computing Machinery, New York, NY, USA, 5540–5544. https://doi.org/10.1145/3746252.3760832

### BibTeX Citation
```bibtex
@inproceedings{10.1145/3746252.3760832,
  author = {\v{Z}id, \v{C}en\v{e}k and Kord\'{\i}k, Pavel and Alves, Rodrigo},
  title = {Active Recommendation for Email Outreach Dynamics},
  year = {2025},
  isbn = {9798400720406},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  url = {https://doi.org/10.1145/3746252.3760832},
  doi = {https://doi.org/10.1145/3746252.3760832},
  booktitle = {Proceedings of the 34th ACM International Conference on Information and Knowledge Management},
  pages = {5540–5544},
  numpages = {5},
  keywords = {email outreach, reinforcement learning, shallow autoencoder},
  location = {Seoul, Republic of Korea},
  series = {CIKM '25}
}
```

![Global open rate and distribution of per-user open rates.](./assets/user_open_rate_hist.png)
Global open rate and distribution of per-user open rates.

## Time to Open (TTO)
Time-to-open is heavy-tailed: while the median is about 6.4 hours, most opens occur within a week. Specifically, 
93.0\% of opens arrive within 7 days, so 7.0\% arrive later than 7 days. The plots below are truncated at 7 days to 
emphasize the main mass of the distribution. The CDF and histogram are shown in Figure~\ref{fig:tto}.

![Distribution of time-to-open for opened emails.](./assets/time_to_open_hist.png)
Distribution of time-to-open for opened emails.

![CDF of time-to-open for opened emails.](assets/time_to_open_cdf.png)
CDF of time-to-open for opened emails.

The heavy-tailed TTO suggests robust objectives and appropriate censoring strategies. The two user segments motivate 
segment-aware priors and exploration strategies; mailshot-level heterogeneity motivates per-mailshot features or random effects.

## Dataset Versions
The current version of the dataset contains 12 months of data (2024-04 -- 2025-03). Future dataset might
include additional months of data. The data collection is still ongoing.


## Acknowledgements
Čeněk Žid's research was supported by the Grant Agency of the Czech Technical University (SGS20/213/OHK3/3T/18). 
We warmly thank *Mailprofiler* for providing the dataset for this research.


<p align="center">
  <a href="https://fit.cvut.cz/en" target="_blank">
    <img src="assets/logo-fit-en-modra.jpg" alt="FIT CTU" height="60"/>
  </a>
  &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
  <a href="https://xcampaign.info/switzerland-en/" target="_blank">
    <img src="assets/Xcampaign_logo.svg" alt="XCampaign" height="60"/>
  </a>
  &nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
  <a href="https://www.recombee.com/" target="_blank">
    <img src="assets/recombee_logo.png" alt="Recombee" height="60"/>
  </a>
</p>