Tenrec Large-Scale Multipurpose Recommender System Benchmark Dataset Explained: 5 Million Users, 140 Million Interactions, 4 Scenarios

Why do you need a recommender system benchmark dataset that's closer to reality?

When many people start recommender systems research, the first thing they think of is a smaller dataset that contains only a single type of positive feedback. Such datasets aren't unusable, but the practical value of models in real systems tends to be discounted.

Tenrec was created to fill this gap. It publicly provides large-scale, multi-scenario, and diverse user feedback, giving recommendation models an evaluation foundation that's closer to real environments.

What is Tenrec?

Tenrec is a public large-scale multipurpose benchmark dataset for recommender systems, from the paper "Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems," published at NeurIPS 2022 Datasets and Benchmarks Track.

It collects data from two different feeds recommendation apps at Tencent, covering four recommendation scenarios. For those who need large-scale, multi-scenario, reproducible experiments, this is a choice worth getting to know first.

Tenrec's five core features

1. Large-scale: About 5 million users and 140 million interactions, far exceeding many common benchmark datasets.
2. Both positive and negative feedback: Not just positive feedback, but also real negative feedback, unlike single-class recommendation datasets that have only one type of signal.
3. Cross-scenario overlap: The four different scenarios have overlapping users and items, which can be used to study cross-domain and transfer.
4. Multiple types of positive feedback: Includes clicks, likes, shares, follows, and more, closer to the diverse interactions in real products.
5. Additional features: Besides user ID and item ID, it also includes other features; the source also mentions read_time and other reading duration information.

Tenrec's four recommendation scenarios

Tenrec's data comes from two different feeds recommendation apps at Tencent, covering four recommendation scenarios in total.

The source mentions that one scenario is QK-video; based on the public abstract, production system names have been anonymized, and the remaining scenario names and details should be based on the paper and official data page.

This cross-scenario design allows the same dataset to support more diverse recommendation tasks, rather than serving only a single application.

What tasks can Tenrec be used to evaluate?

The source states that Tenrec can be used to benchmark ten diverse recommendation tasks and covers most common recommendation scenarios.

Among them, cold start is an important and not yet fully solved challenge in recommender systems. Tenrec's data scale and additional features make it suitable for this type of research.

It releases the dataset and code, aiming to promote reproducibility and advance new recommendation research.

Who is Tenrec suitable for?

• Recommender systems researchers: Those who need a large-scale, multi-scenario, multi-feedback public benchmark.
• Students and experimental teams: Those who want to validate models with real data, not just toy datasets.
• Those who need reproducible experiments: Tenrec provides public data and code, making it easy to compare with paper results.
• Those studying cross-domain recommendation: The four scenarios have overlapping users and items, suitable for extending cross-scenario topics.

How to access Tenrec and how to cite it?

Tenrec is a public dataset. The source mentions that the dataset and code have been released, and related information can be found on arXiv, OpenReview, and NeurIPS 2022 proceedings.

The official data page is https://tenrec0.github.io/, arXiv page is https://arxiv.org/abs/2210.10629.

If used for research, the citation format provided by the source is:

@article{yuan2022tenrec, title={Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems}, author={Yuan, Guanghu and Yuan, Fajie and Li, Yudong and Kong, Beibei and Li, Shujie and Chen, Lei and Yang, Min and Yu, Chenyun and Hu, Bo and Li, Zang and others}, journal={arXiv preprint arXiv:2210.10629}, year={2022} }

4 things to know before using Tenrec

1. Read the paper before downloading: The four scenarios, feature fields, and task splitting methods should all be based on the paper's definitions.
2. Note production system anonymization: The source states that the two production system names have been anonymized; do not speculate or claim which platforms they correspond to.
3. Comply with data licensing and citation requirements: Public data does not mean unrestricted use; you still need to check the official license and citation guidelines.
4. Don't just look at scale: Large scale is an advantage, but scenarios, feedback types, and feature design are the key factors determining whether it suits your topic.

Frequently Asked Questions

Is Tenrec a recommender system tool?

No. Tenrec is a public recommender system benchmark dataset, mainly used for training and evaluating recommendation models, not a functional tool for direct operation.

Does Tenrec only have positive feedback?

No. The source explicitly mentions that it includes real negative feedback, as well as multiple types of positive feedback such as clicks, likes, shares, and follows.

Is Tenrec suitable for cold start research?

The source lists cold start as an important and not yet fully solved challenge, and Tenrec's scale and additional features make it suitable for related experiments.

Can Tenrec be used in commercial products?

This should be based on the official license and terms of use. Public academic datasets usually have citation and usage scope requirements; read the instructions carefully before use.

Summary

Tenrec is a rare public benchmark dataset in the recommender systems field that simultaneously has large scale, multiple scenarios, and both positive and negative feedback.

If you are looking for a dataset that can support ten diverse recommendation tasks and facilitate reproducible research, Tenrec is worth getting to know starting from the paper and official data page.

Overall, it is not a tool for "trying out," but infrastructure for "validating recommendation models."

References

• 2210.10629 Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems
• Tenrec
• Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems
• Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems | OpenReview
• (PDF) Tenrec: A Large-scale Multipurpose Benchmark Dataset for Recommender Systems