Paper deep dive
NewsTorch: A PyTorch-based Toolkit for Learner-oriented News Recommendation
Rongyao Wang, Veronica Liesaputra, Zhiyi Huang
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%
Last extracted: 4/18/2026, 1:36:01 AM
Summary
NewsTorch is a PyTorch-based, learner-oriented toolkit designed to facilitate research in neural news recommendation. It provides a modular, decoupled framework with a GUI, supporting dataset preprocessing, model training, and evaluation for deep learning, GNN-based, and LLM-based recommendation models.
Entities (5)
Relation Signals (3)
NewsTorch → builton → PyTorch
confidence 100% · We propose a PyTorch-based news recommendation toolkit called NewsTorch
NewsTorch → supports → MIND
confidence 95% · Various popular news recommendation datasets have been employed in our toolkit, such as MIND
NewsTorch → supports → EB-NeRD
confidence 95% · Various popular news recommendation datasets have been employed in our toolkit, such as ... EB-NeRD
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:News recommender systems are devised to alleviate the information overload, attracting more and more researchers' attention in recent years. The lack of a dedicated learner-oriented news recommendation toolkit hinders the advancement of research in news recommendation. We propose a PyTorch-based news recommendation toolkit called NewsTorch, developed to support learners in acquiring both conceptual understanding and practical experience. This toolkit provides a modular, decoupled, and extensible framework with a learner-friendly GUI platform that supports dataset downloading and preprocessing. It also enables training, validation, and testing of state-of-the-art neural news recommendation models with standardized evaluation metrics, ensuring fair comparison and reproducible experiments. Our open-source toolkit is released on Github: this https URL.
Tags
Links
- Source: https://arxiv.org/abs/2604.14510v1
- Canonical: https://arxiv.org/abs/2604.14510v1
Trouble viewing inline? Open PDF directly →
Full Text
10,731 characters extracted from source content.
Expand or collapse full text
NewsTorch: A PyTorch-based Toolkit for Learner-oriented News Recommendation Rongyao Wang 1 , Veronica Liesaputra 1 , Zhiyi Huang 1 1 School of Computing, University of Otago, Dunedin 9016, New Zealand rongyao.wang@postgrad.otago.ac.nz, veronica.liesaputra@otago.ac.nz, zhiyi.huang@otago.ac.nz Abstract News recommender systems are devised to alleviate the in- formation overload, attracting more and more researchers’ attention in recent years. The lack of a dedicated learner- oriented news recommendation toolkit hinders the advance- ment of research in news recommendation. We propose a PyTorch-based news recommendation toolkit called New- sTorch, developed to support learners in acquiring both conceptual understanding and practical experience. This toolkit provides a modular, decoupled, and extensible frame- work with a learner-friendly GUI platform that supports dataset downloading and preprocessing. It also enables training, validation, and testing of state-of-the-art neural news recommendation models with standardized evalua- tion metrics, ensuring fair comparison and reproducible ex- periments. Our open-source toolkit is released on Github: https://github.com/whonor/NewsTorch. Introduction Background: News recommender system is one type of recommender systems, which is able to recommend inter- ested news content to users and reduce the impact of the information overload problem. Various news online plat- forms have implemented news recommender systems, such as Bing News, CNN and BBC. News recommendation has become a vital role in influencing users’ reading behaviours in the era of artificial intelligence (Wang et al. 2024). Motivation: In recent years, deep learning technology has achieved promising advancements, including generative large language models (LLMs) and artificial intelligence agents. Some superior deep learning-based methods (Wu et al. 2019; Wang et al. 2022, 2023; Wu et al. 2021) are uti- lized in news recommender systems. News recommendation has attracted increasing attention from researchers, result- ing in a large number of publications in recent years (Wu et al. 2024). However, studying news recommendation re- mains challenging for beginners due to complex data prepa- ration, difficulties in reproducing code, and heterogeneous evaluation frameworks. Challenge: The news recommendation toolkit is a valu- able resource that facilitates experimental research by pro- viding a fair evaluation platform across different baselines. Copyright © 2026, Association for the Advancement of Artificial Intelligence (w.aaai.org). All rights reserved. Most open-source news recommendation libraries (Iana, Glava ˇ s, and Paulheim 2023) devise a unified and highly con- figurable framework for researchers built on a high-level library such as PyTorch-Lightning 1 , Hydra(Yadan 2019). While existing high-level libraries support reproducible re- search and comprehensive experiments, they provide limited support for learners. This motivates the design of a learner- friendly toolkit for news recommendation studies. There- fore, we propose NewsTorch, a framework designed to fa- cilitate model reproduction, dataset preprocessing, and ex- perimental operations with ease for learners. The essential differences between NewsTorch and other popular news rec- ommendation libraries are demonstrated in Table 1. Contribution: (1) In this paper, we propose a learner- oriented news recommendation toolkit named NewsTorch, which includes a unified framework and a learner-friendly GUI (Graph User Interface) to configure, train, and evaluate all baselines. (2) We reproduce a wide range of neural news recommendation models, including deep learning-based models, GNN-based models, and LLM-based news recom- mendation models. Moreover, NewsTorch allows learners to flexibly customize models and datasets via an independent configuration mechanism. (3) To the best of our knowledge, NewsTorch is the first learner-oriented news recommenda- tion toolkit to support both GNN-based and LLM-based models within a decoupled framework. Overview of NewsTorch In this section, we introduce the framework and compo- nents of NewsTorch. The framework is illustrated in Figure 1, which is composed of different functional components: Dataset Preparer, Experiment Controller, Model Man- ager and Web GUI. Dataset Preparer. This component is devised to down- load and process different news recommendation datasets, including Dataset Downloader and Dataset Corpus Pro- cessor. The Dataset Downloader is responsible for acquir- ing datasets, whereas Dataset Corpus Processor enables the transformation of diverse news recommendation datasets into a unified format. Various popular news recommenda- tion datasets have been employed in our toolkit, such as 1 https://lightning.ai/pytorch-lightning arXiv:2604.14510v1 [cs.IR] 16 Apr 2026 Experiment Controller Data Preparer Model Manager Dataset Downloader Dataset Corpus Processor Configuration Director Experiment Runner News EncodersUser Encoders Neural Network Layers Special modules Downloading (e.g., dataset, embeddings) Loading (e.g., tsv, json, pkl) Parsing (e.g., MIND, EB-NeRD) Loading (e.g., tsv, json, pkl) Splitting (e.g., train, dev, test) Sampling (e.g., negative, positive) Unzip (e.g., zip) Setting experiments (e.g., batch_size, epoch) Configuring models (e.g., dimension, layer) Training Validating Testing Encoding (e.g., title, categories, abstract) Encoding (e.g., behaviors) Projection (e.g., MLP) Convolution (e.g., CNN) Attention (e.g., transformers) Tool (e.g., KG, TF-IDF, LLMs) Evaluation (e.g., AUC, MRR) Logging (e.g., Weights & Biases) Logging Figure 1: The framework of the NewsTorch toolkit. MIND(Wu et al. 2020) and EB-NeRD(Kruse et al. 2024), which can be easily utilized by learners in their experiments. Experiment Controller. This component enables users to independently configure models and experiments and to initiate experimental runs, comprising Configuration Di- rector and Experiment Runner. Specifically, Configura- tion Director is utilized to assist users in customizing pa- rameters with an independent configuration mechanism, us- ing YAML files to configure each model within a dedi- cated directory. Experiment Runner is designed to exe- cute different neural news recommendation models, sup- porting independent training, validation, and testing. In or- der to record experiments’ status and results, we introduce Weights&Biases 2 into Experiment Runner. Model Manager. We design this component to manage all models’ codes, including news encoders, user encoders, various neural network layers, and special modules. Suffi- cient and diverse news recommendation models are covered in our toolkit, such as deep learning-based models, GNN- based models, and LLM-based models. Moreover, adding new models to this framework is straightforward due to its modular and decoupled design. Web GUI. This provides various functions that assist learners in directly downloading and processing datasets. After preprocessing, they can click the ’start’ button to train, validate, and test models. 2 https://wandb.ai/site FeatureNews- Recommendation a NewsRecLib b NewsTorch ModelsDL c DLDL,GNN,LLM Multi-Dataset✗✓ Active✗✓ Multi-GPU✗✓ GUI✗✓ a https://github.com/yusanshi/news-recommendation b https://github.com/andreeaiana/newsreclib c Deep learning-based methods Table 1: Compact comparison of news recommendation li- braries. Conclusion NewsTorch is a neural news recommendation toolkit to sup- port learners in acquiring both conceptual understanding and practical experience. It encourages a fair experimental en- vironment and a unified framework with a learner-friendly GUI. With a modular and extensible architecture, the toolkit allows learners to flexibly customize models and datasets via an independent configuration mechanism. Moreover, New- sTorch supports advanced functionalities, including multi- GPU training, experiment tracking, the implementation of diverse models, and the processing of multiple datasets. In the future, we plan to explore more advanced functions and continue updating our toolkit. New state-of-the-art news rec- ommendation models and datasets will be incorporated into NewsTorch. Acknowledgments This work is supported by the University of Otago - China Scholarship Council Doctoral Scholarship, grant number 202308370189. We would also like to thank the anonymous AAAI reviewers for their positive feedback and helpful sug- gestions. References Iana, A.; Glava ˇ s, G.; and Paulheim, H. 2023. NewsRecLib: A PyTorch-Lightning Library for Neural News Recommen- dation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing: System Demon- strations, 296–310. Kruse, J.; Lindskow, K.; Kalloori, S.; Polignano, M.; Pomo, C.; Srivastava, A.; Uppal, A.; Andersen, M. R.; and Frellsen, J. 2024. EB-NeRD a large-scale dataset for news recom- mendation. In Proceedings of the Recommender Systems Challenge 2024, 1–11. Wang, R.; Wang, S.; Lu, W.; and Peng, X. 2022. News Recommendation Via Multi-Interest News Sequence Mod- elling. In ICASSP 2022-2022 IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), 7942–7946. Wang, R.; Wang, S.; Lu, W.; Peng, X.; Zhang, W.; Zheng, C.; and Qiao, X. 2023. Intention-aware user modeling for per- sonalized news recommendation. In International Confer- ence on Database Systems for Advanced Applications, 179– 194. Springer. Wang, S.; Zhang, X.; Wang, Y.; and Ricci, F. 2024. Trust- worthy recommender systems. ACM Transactions on Intel- ligent Systems and Technology, 15(4): 1–20. Wu, C.; Wu, F.; Ge, S.; Qi, T.; Huang, Y.; and Xie, X. 2019. Neural News Recommendation with Multi-head Self- attention. In Proceedings of the 2019 conference on empiri- cal methods in natural language processing and the 9th in- ternational joint conference on natural language processing (EMNLP-IJCNLP), 6390–6395. Wu, C.; Wu, F.; Qi, T.; and Huang, Y. 2021. Empowering News Recommendation with Pre-trained Language Models. In Proceedings of the 44th international ACM SIGIR confer- ence on research and development in information retrieval, 1652–1656. Wu, F.; Qiao, Y.; Chen, J.-H.; Wu, T., Chuhan Qi; Lian, J.; Liu, D.; Xie, X.; Gao, J.; and Wu, W. 2020. MIND: A Large- scale Dataset for News Recommendation. In Proceedings of the 58th annual meeting of the association for computa- tional linguistics, 3597–3606. Wu, L.; Zheng, Z.; Qiu, Z.; Wang, H.; Gu, H.; Shen, T.; Qin, C.; Zhu, C.; Zhu, H.; Liu, Q.; et al. 2024. A survey on large language models for recommendation. World Wide Web, 27(5): 60. Yadan, O. 2019. Hydra - A framework for elegantly config- uring complex applications. Github.