RSPapers: a curated survey of recommender systems research

A 17-category reading list for recommender system research

RSPapers is a curated repository of academic papers on recommender systems, organized by research domain and technique. The list spans from foundational work on collaborative filtering and matrix factorization to recent developments in large language models and agentic systems for recommendations. Each category groups papers by the specific problem they address: how to overcome the cold-start problem when you have no history for a user, how to use knowledge graphs as side information to improve recommendations, how to explain recommendations to users, how to handle exploration and exploitation trade-offs, and others. The repository includes 17 major categories: Tutorials, Surveys, General RS, Social RS, Deep Learning-based RS, Cold Start, POI RS, Efficient RS, Exploration and Exploitation, Explainability, CTR Prediction, Knowledge Graph for RS, Review-based RS, Conversational RS, Industrial RS, Privacy and Security RS, and LLM for RS, plus a new category on Agentic RS. Within each category, papers are listed with author, title, and publication venue and year.

Browsing RSPapers on GitHub

RSPapers lives as a public repository on GitHub at hongleizhang/RSPapers. To start reading, visit the repository and open the README in your browser:

https://github.com/hongleizhang/RSPapers

The README lists all 17 categories with links to each one. Click on a category that matches your interest, such as 04-Deep_Learning_based_RS or 16-LLM_for_RS, and you will see papers listed with authors, titles, conferences and publication years. Each category is a markdown file in the repository. If you want to clone the repository locally, you can run `git clone https://github.com/hongleizhang/RSPapers.git` to download it.

How to navigate the paper list and contribute

RSPapers is organized with each category as a top-level markdown file. The repository also maintains a section called 00-Latest_Papers with new papers added recently. Papers are submitted as pull requests using a standard template: author names, paper title, conference or journal, and publication year. This means the list is community-maintained and grows when researchers and practitioners propose additions. The README specifies the exact format for pull requests, making it straightforward to add papers you have found or written. The last push was on 2026-03-12, indicating that the list was actively maintained as of six months ago. New papers in areas like LLM for RS and Agentic RS reflect the field's evolution.

Tutorials from top-tier venues and comprehensive surveys

The Tutorials category contains presentations and courses given by prominent researchers at major conferences like Recsys, ICML, SIGIR, WWW, IJCAI, CIKM and WSDM. These include foundational topics like recommender problems for web applications, the user experience of recommendations, and cross-domain recommendations. Recent tutorials cover deep learning for recommendations, fairness in recommendations, graph neural networks, and self-supervised learning. Each tutorial entry names the author, the tutorial title and the conference. The Surveys category provides comprehensive overviews of specific domains: hybrid recommender systems, social recommender systems, and deep-learning-based recommender systems. Surveys are the entry point for understanding a problem area; they cite the foundational papers and explain the landscape of solutions.

Foundational and domain-specific recommendation techniques

The General RS category covers classic models and practical theory, including collaborative filtering, matrix factorization, and early deep-learning approaches. The Deep Learning-based RS section contains papers on neural networks for recommendations, from early work with autoencoders and restricted Boltzmann machines to modern graph neural networks and transformer-based approaches. Social RS papers describe how to use trust relationships and social information to improve predictions. The Cold Start problem is addressed in its own category, with papers on how to make good recommendations for new users with no history or new items with no ratings. POI RS focuses on location-based recommendations. Efficient RS covers scalable algorithms for large-scale systems. CTR Prediction papers address the specific problem of predicting whether a user will click on an item, a critical component in many industry recommendation pipelines. Knowledge Graph for RS explores using structured knowledge as side information to improve recommendations and combat data sparsity.

Modern approaches: explanations, conversations, and language models

Explainability in RS papers address the problem of not just recommending items, but explaining why you are recommending them. Conversational RS covers natural language interaction with recommendation systems, allowing users to refine their preferences through dialogue. Review-based RS uses text reviews as a signal for recommendations. Privacy and Security RS papers address threats to user data and fairness concerns in recommendation algorithms. The Industrial RS category contains papers on best practices from companies operating large-scale recommender systems. Two newer categories reflect recent developments in the field: LLM for RS covers large language models applied to recommendations, including prompt-based recommendation, and Agentic RS covers LLM-based agents that can interact with users and systems to make recommendations. These categories show the field's evolution from traditional matrix-based algorithms toward foundation models.

Exploration and exploitation, and explainable recommendations

Exploration and Exploitation papers address a core problem in recommendations: whether to recommend items similar to a user's history (exploitation) or new items that might broaden their interests (exploration). This trade-off matters in interactive systems where user feedback changes over time. Explainability papers go beyond the accuracy of recommendations and ask how to help users understand why an item was recommended. This is critical in high-stakes domains like job recommendations or content moderation. Papers in this area cover rule-based explanations, feature importance, and counterfactual reasoning.

Using RSPapers vs. reading papers directly vs. implementation resources

RSPapers serves a different purpose than a paper database like arXiv or Google Scholar. It is a curated list with human judgment about which papers matter, organized by research problem rather than chronology or venue. arXiv lets you search by keyword and see all papers on a topic, but offers no guidance on which are foundational and which are incremental. Google Scholar provides citation counts and impact metrics. RSPapers offers structure: it groups papers by the specific problem you are trying to solve, meaning you can find relevant work without knowing keywords or venue names. The trade-off is that RSPapers is static. It does not rank papers by impact or recency within each category, and it does not discuss implementation choices or compare approaches. You will need to read the papers themselves to understand trade-offs. For hands-on learning, implementation libraries and tutorials matter more than papers alone; the list includes tutorials but does not cover code repositories or libraries like Surprisal, LensKit or implicit.

Editorial conclusion

RSPapers is essential for anyone building recommender systems who needs to understand the academic foundation of the field. Researchers implementing recommendation algorithms should start here; so should engineers in industry evaluating approaches for specific problems. The repository organizes papers by 17 research areas, from foundational surveys to recent work on agentic systems and privacy. Start by reading a survey paper in your domain (cold-start for new-user problems, knowledge graphs if you have structured metadata, LLMs if you are considering language-based recommendations). Then follow the paper citations to go deeper. The main constraint is that this is a static list without discussion of implementation choices or comparisons between approaches; you will need to read the papers themselves to understand trade-offs.

Frequently asked questions

Official sources

1. hongleizhang/RSPapers on GitHub
2. Issues
3. License: MIT
4. README