tim-rec / README.md
shba93's picture
Update README.md
6ec7c67 verified

TIM Rec: Sparse Feedback on Multi-Item Upselling Recommendations in an Industrial Dataset of Telco Calls

[RECSYS 2025] TIM Rec: Sparse Feedback on Multi-Item Upselling Recommendations in an Industrial Dataset of Telco Calls

Authors: Alessandro Sbandi, Federico Siciliano, Fabrizio Silvestri

GitHub: https://github.com/ashba93/tim-rec/tree/main

Citation

If you use this code or find it helpful for your research, please cite:

@inproceedings{sbandi2025tim,
  title={TIM-Rec: Explicit Sparse Feedback on Multi-Item Upselling Recommendations in an Industrial Dataset of Telco Calls},
  author={Sbandi, Alessandro and Siciliano, Federico and Silvestri, Fabrizio},
  booktitle={Proceedings of the Nineteenth ACM Conference on Recommender Systems},
  pages={865--873},
  year={2025}
}

Overview

This dataset introduces a real-world telecommunications upselling dataset with multi-item recommendations and sparse feedback. The dataset captures customer interactions from a real customer care service, where multiple items can be recommended in a single session, and only a small fraction of offers are accepted. Our goal is to provide a resource for evaluating Learning-to-Rank (LTR) models and recommendation systems in real-world settings.

Key Features

  • Real-World Telco Data: Collected from customer care interactions in a live telecommunications setting.
  • Multi-Item Recommendations: Each session contains multiple recommendations, reflecting real-world upselling scenarios.
  • Sparse Feedback: A low acceptance rate, emphasizing the challenge of learning from limited positive signals.
  • Benchmark Baselines: We provide results from various recommendation models, ranging from classical approaches to deep learning-based solutions.
  • Rich Feature Set: Includes user contextual features, recommendation metadata, and interaction timestamps.

Dataset Statistics

Characteristic Value
Total interactions 1494061
Time range ∼6 months
Acceptance sparsity (%) ∼5%
Avg. recommendations per session ∼2.4

For more details on the dataset schema and preprocessing, refer to the documentation.

Benchmark Models

We evaluate several Learning-to-Rank and recommendation models, including:

  • Traditional Models: Matrix Factorization Algorithm
  • Neural Models: Neural Network based approaches
  • Graph-Based Models: GNN-based recommendation approaches

Results and evaluation metrics such as Precision@K, NDCG@K, and Recall@K are included in the paper.

Contact

For questions or collaborations, please open an issue or contact us at alessandro.sbandi@gmail.com.


license: cc-by-nc-4.0 size_categories: - 1M<n<10M