RSPapers: A Markdown Reading List for Recommender Systems Research

What RSPapers Actually Is: A Reading List, Not a Toolkit

RSPapers is a single Markdown file that collects paper citations and conference tutorial listings for recommender systems research. The README describes it as a curated list of papers and tutorials about Recommender Systems, and the repository has no code, no build step and no dependencies. The only artefact is the README itself plus a LICENSE file. That matters because the name invites a category error. A reader who lands on the repository expecting a framework, a dataset loader or a baseline implementation will find none of those things. What they will find is an index: author names, paper titles, venue and year, arranged under seventeen numbered headings. The contribution template in the README is a single Markdown bullet of the form author, bolded paper name, venue, year. There is no schema beyond that, no metadata about whether a paper has released code, and no abstraction of the papers into topics or methods. If you want a taxonomy you build it yourself from the headings, which range from 00-Tutorials through 17-Agentic RS. The list is deliberately flat. Each heading is a bucket, and the entries inside it are ordered by nothing in particular that the README states.

The Seventeen Sections and What Each One Assumes You Already Know

The section list doubles as a rough map of the field. 00-Tutorials collects conference tutorials from ICML, RecSys, SIGIR, CIKM, WWW, WSDM and IJCAI, spanning 2011 to 2021 in the portion of the README shown. 01-Surveys points at survey papers on hybrid, social and POI recommenders. 02-General RS holds classic prediction models. 03-Social RS targets trust and social signals used to offset rating sparsity, which the README states directly. 04-Deep Learning-based RS and 05-Cold Start Problem in RS sit next to each other, which is a sensible adjacency since deep models are frequently proposed as a cold-start remedy. 06-POI RS covers location-based social networks. 07-Efficient RS and 08-EE Problem in RS cover training efficiency and the exploration-exploitation trade-off respectively. 09-Explainability on RS is framed around the why question, and the README's own wording is that these papers generate recommendation explanations rather than only producing ranked items. 10-CTR Prediction for RS is positioned as candidate-set elaboration, a narrower framing than treating CTR as a standalone ranking problem. 11-Knowledge Graph for RS, 12-Review based RS, 13-Conversational RS, 14-Industrial RS, 15-Privacy&Security RS, 16-LLM for RS and 17-Agentic RS close the list. The newest three sections, LLM for RS, Agentic RS and Privacy&Security RS, are flagged in the README with New! markers, which is the only signal in the document about which parts have grown recently.

How to Use the List Without Wasting a Week on Dead Links

The README gives no tooling, so the practical workflow is whatever your Markdown reader supports. Clone the repository and read the file locally, or use the anchor links in the table of contents, which point at GitHub heading fragments such as #agentic-rs and #llm-for-rs. The anchors matter because the file is long enough that scrolling is unpleasant. There is no search index, no BibTeX export and no JSON sidecar. If you need citations in a reference manager you will be transcribing entries by hand, and the README's entry format is inconsistent enough that automated parsing is unreliable: some entries use et al. after a first name, some use a surname, some tutorial entries begin with a hyphen inside the title string as in the Bart et al. privacy tutorial line. That inconsistency is a real cost. A single contributor-facing template exists in the README, but nothing enforces it on merged entries, so the file has drifted. For a reader who wants to skim, the numbered prefixes are the fastest navigation aid. For a reader who wants to build a reading queue, the honest answer is that this repository gives you the raw material and none of the plumbing.

Contributing Is a Pull Request Against the README

The README states that contributions should come as a pull request using the template it provides: an author name, a bolded paper title, the venue and the year. That is the whole contribution process described. There is no CONTRIBUTING file mentioned, no CI configuration referenced, no link checker, and no stated review criteria beyond the template. In practice this means the quality bar is editorial and human. A contributor can add a paper that is off-topic, duplicate an existing entry under a different section, or attach the wrong venue, and nothing in the repository will catch it. For a project of this shape that is normal, but it is worth stating plainly because it determines what you can trust. The presence of a paper in a section means a maintainer or contributor thought it belonged there. It does not mean the paper was reproduced, that its claims survived scrutiny, or that its code, if any, still runs. The README also links a WeChat account and an update badge reading weekly, which are the only maintenance signals visible in the supplied material. Neither tells you which sections were touched.

Where a Static Paper List Stops Being the Right Tool

The failure mode is straightforward. A Markdown list has no mechanism for marking a paper as superseded, retracted, or built on assumptions that later work overturned. If a 2016 cold-start method was eclipsed by a 2022 graph approach, both entries sit in the file with equal visual weight unless a human edits the older one. The README's structure gives no place to record that relationship. The same applies to code availability. A reader scanning 04-Deep Learning-based RS cannot tell from the entry format whether a paper shipped an implementation, which is often the deciding factor for an engineer choosing what to reproduce. There is also a currency problem specific to the newer sections. LLM for RS and Agentic RS are fast-moving areas where a list updated on a weekly cadence, as the badge claims, may still lag the arXiv firehose by months. The README marks these sections as new additions, which suggests they were appended rather than continuously curated. None of this makes the list bad. It makes it a snapshot with an editorial layer, and snapshots need to be read as such. If your question is which method to deploy next quarter, this repository will not answer it.

RSPapers Versus Awesome Recommender Systems and Papers with Code

The closest comparison in kind is the Awesome Recommender Systems list, which follows the same awesome-list convention: a Markdown index under a permissive licence. The difference is scope and framing. Awesome-style lists typically mix papers, libraries, datasets and tutorials in one flat index, which makes them useful as a starting directory but noisy when you want only literature. RSPapers is narrower and more academic: it is papers and conference tutorials, organised by research subarea rather than by resource type. That makes it better for a literature review and worse for finding a library to install. Papers with Code takes the opposite approach, attaching implementations and reported results to papers and indexing them in a database rather than a file. Its weakness is coverage bias toward papers that released code, and its strength is that you can filter by task and dataset. RSPapers has neither filter nor database, but it also does not exclude theory papers that never shipped an implementation. If your goal is to find a reproducible baseline, Papers with Code is the more direct route. If your goal is to understand how a subarea developed, RSPapers' section structure gives you a reading order that a code-centric index does not.

Licence, Maintenance and the Cost of Keeping a List Alive

The repository is MIT licensed, which is permissive for the text as published. That covers the list itself, not the papers it cites. Each linked paper carries its own publisher or author terms, and several venues referenced in the README, including RecSys, SIGIR, CIKM, WWW and WSDM, are typically paywalled or ACM-licensed. The MIT licence on RSPapers grants you nothing with respect to those PDFs, and the README does not claim otherwise. Anyone planning to redistribute the list, for example inside a course syllabus or an internal wiki, should treat the citations as facts and the surrounding text as MIT-licensed, without assuming the same for the underlying works. This is a description of the licence text, not legal advice. On maintenance, the costs are the ones you would expect for a manually edited file: review time per pull request, periodic link rot, and the slow accumulation of drift in entry formatting. The repository has no automated checks that the supplied material mentions, so every one of those costs lands on a human. The last push timestamp is recent, which is consistent with the weekly update badge, but the README does not record per-section edit dates, so you cannot tell from the file alone which parts are fresh.

Who Should Adopt RSPapers and What to Check First

Adopt it if you are starting a literature review in a specific recommender systems subarea and want a pre-filtered entry point. The seventeen-section split means you can go straight to, say, 15-Privacy&Security RS or 13-Conversational RS and skip the rest. Adopt it if you are assembling a reading group syllabus and want a defensible starting set that a maintainer already vetted for topical fit. Do not adopt it if you need runnable code, benchmark comparisons, or an up-to-date view of a fast-moving area; the format cannot carry any of those. Do not adopt it as a citation source without opening each paper, because the entry format does not distinguish a peer-reviewed paper from a workshop note or a tutorial abstract. Before you commit to it, check two things. First, open the LLM for RS and Agentic RS sections and see whether the most recent entries match the current state of those areas; that tells you how live the curation is. Second, check the commit history for edits to the sections you care about, since the README itself does not date its entries. If both look current for your subarea, the list is doing its job. If the sections you need have not moved in a year, you are reading an archive, and the MIT licence will not make it a current one.

Editorial conclusion

RSPapers suits graduate students, applied researchers and engineers entering a recommender systems subarea who need a starting bibliography and are willing to verify each paper themselves. It does not suit anyone expecting runnable code, benchmark numbers or maintained annotations. Before relying on it, open the Agentic RS and LLM for RS sections to gauge how current the coverage is, and check the commit history to confirm the weekly update claim still holds for the subareas you care about.

Official sources

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