sm-rs 0.1.0__tar.gz

sm_rs-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,24 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.11", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+          cache-dependency-path: pyproject.toml
+      - name: Install
+        run: python -m pip install -e ".[dev]"
+      - name: Test
+        run: python -m pytest -q

sm_rs-0.1.0/.gitignore

@@ -0,0 +1,11 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+build/
+dist/
+.eggs/
+.pytest_cache/
+.ruff_cache/
+.venv/
+venv/
+.DS_Store

sm_rs-0.1.0/CITATION.cff

@@ -0,0 +1,46 @@
+cff-version: 1.2.0
+message: "If you use the SM-RS dataset or benchmark, please cite BOTH papers below."
+title: "SM-RS benchmark"
+abstract: >-
+  Data loaders and canonical task evaluators for SM-RS, the single- and
+  multi-objective recommendations dataset linking self-declared propensities with
+  contextual impressions, item selections, and perceived quality.
+# Authors of *this software* (the benchmark code). The dataset's authors are in
+# the citations below — please cite the papers.
+authors:
+  - family-names: Dokoupil
+    given-names: Patrik
+repository-code: "https://github.com/pdokoupil/SM-RS"
+license: MIT
+type: software
+# Primary citation: the journal (SM-RS 2.0) superset.
+preferred-citation:
+  type: article
+  title: "SM-RS 2.0: User-perceived Qualities of Single- and Multi-Objective Recommender Systems"
+  authors:
+    - family-names: Dokoupil
+      given-names: Patrik
+    - family-names: Peska
+      given-names: Ladislav
+  year: 2026
+  doi: "10.1145/3754459"
+  journal: "ACM Transactions on Recommender Systems"
+  publisher:
+    name: "Association for Computing Machinery"
+  volume: 4
+  issue: 3
+# Also cite the original SM-RS (SIGIR'24).
+references:
+  - type: conference-paper
+    title: "SM-RS: Single- and Multi-Objective Recommendations with Contextual Impressions and Beyond-Accuracy Propensity Scores"
+    authors:
+      - family-names: Dokoupil
+        given-names: Patrik
+      - family-names: Peska
+        given-names: Ladislav
+      - family-names: Boratto
+        given-names: Ludovico
+    year: 2024
+    doi: "10.1145/3626772.3657863"
+    collection-title: "Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval"
+    collection-type: proceedings

sm_rs-0.1.0/DATASET_CARD.md

@@ -0,0 +1,212 @@
+---
+# Hugging Face dataset card frontmatter (draft). Paste as README.md in the HF dataset repo.
+pretty_name: "SM-RS: Single- and Multi-Objective Recommendations"
+license: cc-by-4.0
+language: [en]
+task_categories: [other]
+tags:
+  - recommender-systems
+  - multi-objective
+  - beyond-accuracy
+  - impressions
+  - propensity
+  - user-study
+  - benchmark
+configs:
+  - config_name: behaviors
+    data_files: behaviors.parquet
+  - config_name: propensities
+    data_files: propensities.parquet
+  - config_name: objective_perceptions
+    data_files: objective_perceptions.parquet
+  - config_name: criteria_values
+    data_files: criteria_values.parquet
+  - config_name: comparative_diversity
+    data_files: comparative_diversity.parquet
+  - config_name: users
+    data_files: users.parquet
+  - config_name: movies
+    data_files: movies.parquet
+  - config_name: books
+    data_files: books.parquet
+---
+
+# SM-RS — Single- and Multi-Objective Recommendations
+
+**SM-RS 2.0** — the extended version of SM-RS (SIGIR'24), and a **superset** of v1.
+To our knowledge the only public RecSys dataset linking users' **self-declared
+propensities** toward beyond-accuracy objectives (relevance, diversity, novelty,
+exploration) with **contextual impressions**, item selections, and **explicit
+perceived-quality** judgments — across two domains (movies, books).
+
+> **If you use this dataset, please cite BOTH papers** (SM-RS 2.0, TORS 2026; and
+> SM-RS, SIGIR 2024) — see [Citation](#citation).
+
+- 📦 **Benchmark code (loaders + canonical evaluators):** https://github.com/pdokoupil/SM-RS
+- 🧊 **Reproducibility archive:** OSF — https://osf.io/wsakx (v2), https://osf.io/hkzje (v1)
+
+## Quick start
+
+```python
+from datasets import load_dataset
+behaviors = load_dataset("pdokoupil/SM-RS", "behaviors")
+# or use the benchmark package: pip install sm-rs
+```
+
+## What's inside
+
+### Core tables
+| Config | Rows | Description |
+|---|---|---|
+| `behaviors` | ~10k | impression logs: `userId`, `timestamp`, `history`, `impressions` (`itemId`/`selected`/`timestamp`), `mors`, `iteration` |
+| `propensities` | — | per-user self-declared `relevance` / `diversity` / `novelty` / `exploration` propensities (+ `userId`, `timestamp`) |
+| `objective_perceptions` | — | per-impression perceived `relevance`/`diversity`/`novelty`/`exploration` (+ `impressionsIds`) |
+| `criteria_values` | — | computed objective values per impression |
+| `comparative_diversity` | — | triplet judgments (`list1`/`list2`/`list3`, `perceived_diversities`, `domain`) |
+| `users` | — | demographics + `domain` (movies/books) |
+| `movies`, `books` | — | item id maps |
+
+### Derived matrices — recomputed locally, **not** hosted here
+Scoring recommendation-list tasks (2, 3) needs item-item (relevance), distance
+(diversity), and popularity (novelty) matrices. These are **not distributed** —
+neither here nor on this dataset — because they derive from third-party catalogs
+(MovieLens, goodbooks-10k) whose terms restrict redistribution, and because they'd
+add gigabytes. Instead the [benchmark package](https://github.com/pdokoupil/SM-RS)
+**recomputes** them from a rating matrix you build from your own download of those
+public datasets (`smrs.derived`: EASE^R item-item, 1−cosine distance, frequency
+popularity — all deterministic). Bit-exact originals are archived on OSF.
+
+## Tasks (benchmark)
+
+Six tasks share this dataset; each has a canonical `evaluate()` in the
+[benchmark repo](https://github.com/pdokoupil/SM-RS) (`smrs.tasks.*`):
+
+| # | Task | Module | Metrics |
+|---|------|--------|---------|
+| 1 | Propensity estimation | `task1_propensity` | MAE · MSE · KL |
+| 2 | Results proportionality | `task2_proportionality` | MAE · KL · wSUM · Pearson ρ |
+| 3 | Selections-aware reranking | `task3_reranking` | nDCG@10 · Precision@5 |
+| 4 | Diversity-metric definition | `task4_diversity` | MAE · MSE · KL |
+| 5 | Perceived quality (5.1 rel / 5.2 div / 5.3 nov / 5.4 ser) | `task5_perceived` | MAE · MSE · Kendall τ |
+| 6 | Satisfaction (6.1 / 6.2) | `task6_satisfaction` | MAE · MSE · Kendall τ |
+
+## Leaderboard
+
+Self-service: run the canonical `evaluate()` and open a PR adding your row (with a
+repro link). The paper baselines below are the rows to beat (from SM-RS 2.0, TORS
+2026, Tables 2–4).
+
+### Task 1 — Propensity estimation · MAE / MSE / KL (↓)
+| Method | MAE | MSE | KL |
+|---|---|---|---|
+| Locally normalized | 0.150 | 0.040 | 0.439 |
+| Globally normalized | 0.149 | 0.039 | 0.271 |
+
+### Task 2 — Results proportionality · MAE↓ / KL↓ / wSUM↑ / ρ↑
+| Method | MAE | KL | wSUM | ρ |
+|---|---|---|---|---|
+| FAI | 0.181 | 0.374 | 0.568 | −0.033 |
+| Probabilistic FAI | 0.141 | 0.242 | 0.568 | 0.318 |
+| Incremental Weighted Average | 0.145 | 0.245 | 0.580 | 0.297 |
+| **RLProp** | **0.114** | **0.181** | 0.549 | **0.356** |
+| GreedyLM | 0.154 | 0.278 | **0.582** | 0.256 |
+| DQNMORS | 0.230 | 0.869 | 0.447 | 0.013 |
+
+### Task 3 — Selections-aware reranking · nDCG@10 / Precision@5 (↑)
+| Method | nDCG@10 | Precision@5 |
+|---|---|---|
+| **Popularity based** | **0.406** | **0.244** |
+| EASE^R | 0.368 | 0.225 |
+| VAE | 0.292 | 0.174 |
+| MultVAE | 0.328 | 0.189 |
+| ImplicitMF | 0.337 | 0.210 |
+| UserKNN | 0.367 | 0.224 |
+| Identity (no reranking) | 0.215 | 0.131 |
+
+### Task 4 — Diversity-metric definition · MAE / MSE / KL (↓)
+| Method | MAE | MSE | KL |
+|---|---|---|---|
+| CF-ILD | 0.470 | 0.316 | 0.279 |
+| CB-ILD | 0.292 | 0.116 | 0.282 |
+| CF-ILMD | 0.409 | 0.246 | **0.272** |
+| **CB-ILMD** | **0.287** | **0.114** | 0.278 |
+| BIN-DIV | 0.294 | 0.127 | 0.288 |
+
+### Task 5 — Perceived quality · MAE↓ / MSE↓ / Kendall τ↑
+
+**5.1 relevance**
+| Method | MAE | MSE | τ |
+|---|---|---|---|
+| Popularity | 0.205 | 0.079 | 0.228 |
+| Relevance-based | 0.191 | 0.063 | **0.263** |
+| Linear Regression | 0.235 | 0.085 | 0.080 |
+| LSTM | **0.189** | **0.061** | 0.022 |
+
+**5.2 diversity**
+| Method | MAE | MSE | τ |
+|---|---|---|---|
+| CF-ILD | 0.270 | 0.116 | 0.171 |
+| CB-ILD | 0.255 | 0.081 | −0.004 |
+| CF-ILMD | 0.227 | 0.074 | 0.146 |
+| CB-ILMD | 0.346 | 0.170 | −0.025 |
+| BIN-DIV | 0.421 | 0.242 | 0.069 |
+| Linear Regression | **0.222** | **0.061** | **0.196** |
+| LSTM | 0.218 | 0.066 | 0.071 |
+
+**5.3 novelty**
+| Method | MAE | MSE | τ |
+|---|---|---|---|
+| Popularity complement | 0.369 | 0.200 | **0.233** |
+| Linear Regression | **0.259** | 0.104 | 0.143 |
+| LSTM | 0.261 | **0.089** | −0.008 |
+
+**5.4 serendipity**
+| Method | MAE | MSE | τ |
+|---|---|---|---|
+| CF-Unexpectedness | 0.303 | 0.142 | **0.128** |
+| CB-Unexpectedness | **0.250** | **0.077** | 0.054 |
+| CF-EASE | 0.283 | 0.117 | 0.068 |
+| CB-EASE | 0.348 | 0.179 | 0.040 |
+| 1−PMI | 0.336 | 0.170 | 0.068 |
+| Linear Regression | 0.270 | 0.103 | 0.039 |
+| LSTM | 0.261 | 0.093 | −0.031 |
+
+### Task 6 — Satisfaction · MAE↓ / MSE↓ / Kendall τ↑
+| Method | MAE | MSE | τ | Subtask |
+|---|---|---|---|---|
+| Popularity based | 0.233 | 0.100 | 0.156 | 6.1 |
+| Relevance based | 0.223 | 0.083 | 0.145 | 6.1 |
+| Linear Regression | 0.255 | 0.102 | 0.045 | 6.1 |
+| LSTM | 0.210 | 0.072 | −0.023 | 6.1 |
+| **LSTM Combined** | **0.138** | **0.035** | **0.515** | 6.2 |
+
+> The `Linear Regression` and popularity/relevance baselines are reproduced by the
+> benchmark package (`examples/reproduce_perceived.py`) — MAE matches the paper.
+
+## What's new vs. SM-RS 1.0 / getting the v1 subset
+SM-RS 2.0 is a superset of the SIGIR'24 release (adds perceived-quality judgments
+and more tasks). The v1 subset is recoverable from the same tables — see the
+benchmark repo for the column mapping.
+
+## Licensing
+Collected study data (this dataset): **CC-BY-4.0** — free to use with attribution
+(cite both papers). The dataset references catalog items **by ID only** and does
+**not** redistribute MovieLens or goodbooks-10k. Derived matrices are recomputed by
+the benchmark from your own copies of those datasets, obtained under **their**
+licenses (MovieLens: no redistribution; goodbooks-10k: per its repository license).
+
+## Citation
+```bibtex
+@article{dokoupil2026smrs2,
+  author  = {Dokoupil, Patrik and Peska, Ladislav},
+  title   = {SM-RS 2.0: User-perceived Qualities of Single- and Multi-Objective Recommender Systems},
+  journal = {ACM Transactions on Recommender Systems}, volume = {4}, number = {3}, year = {2026},
+  doi = {10.1145/3754459}
+}
+@inproceedings{dokoupil2024smrs,
+  author    = {Dokoupil, Patrik and Peska, Ladislav and Boratto, Ludovico},
+  title     = {SM-RS: Single- and Multi-Objective Recommendations with Contextual Impressions and Beyond-Accuracy Propensity Scores},
+  booktitle = {Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
+  series    = {SIGIR '24}, pages = {988--995}, year = {2024}, doi = {10.1145/3626772.3657863}
+}
+```

sm_rs-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Patrik Dokoupil
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

sm_rs-0.1.0/PKG-INFO

@@ -0,0 +1,202 @@
+Metadata-Version: 2.4
+Name: sm-rs
+Version: 0.1.0
+Summary: SM-RS benchmark: data loaders and canonical task evaluators for the single- and multi-objective recommendations dataset.
+Project-URL: Homepage, https://github.com/pdokoupil/SM-RS
+Project-URL: Repository, https://github.com/pdokoupil/SM-RS
+Project-URL: Dataset, https://huggingface.co/datasets/pdokoupil/SM-RS
+Project-URL: Paper (TORS 2026), https://doi.org/10.1145/3754459
+Project-URL: Paper (SIGIR 2024), https://doi.org/10.1145/3626772.3657863
+Author: Patrik Dokoupil
+License: MIT
+License-File: LICENSE
+Keywords: benchmark,beyond-accuracy,dataset,impressions,multi-objective,propensity,recommender-systems
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Requires-Dist: huggingface-hub>=0.20
+Requires-Dist: numpy>=1.21
+Requires-Dist: pandas>=1.3
+Requires-Dist: pyarrow>=12
+Requires-Dist: scikit-learn>=1.0
+Provides-Extra: dev
+Requires-Dist: pytest>=7; extra == 'dev'
+Provides-Extra: lstm
+Requires-Dist: tensorflow>=2.10; extra == 'lstm'
+Description-Content-Type: text/markdown
+
+<div align="center">
+
+# SM-RS
+
+**The single- and multi-objective recommendations benchmark** — self-declared user
+propensities (relevance · diversity · novelty · exploration) linked to contextual
+impressions, item selections, and perceived quality.
+
+<!-- Badges light up once the repo is pushed and the package is published. -->
+[![PyPI](https://img.shields.io/pypi/v/sm-rs.svg)](https://pypi.org/project/sm-rs/)
+[![CI](https://github.com/pdokoupil/SM-RS/actions/workflows/ci.yml/badge.svg)](https://github.com/pdokoupil/SM-RS/actions/workflows/ci.yml)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
+[![Dataset](https://img.shields.io/badge/🤗-dataset-yellow.svg)](https://huggingface.co/datasets/pdokoupil/SM-RS)
+
+</div>
+
+SM-RS is, to our knowledge, the only public recommender-systems dataset linking
+users' **self-declared propensities** toward beyond-accuracy objectives with
+**contextual impressions**, item selections, and **explicit perceived-quality**
+judgments. This repository is the **benchmark code**: data loaders and the
+*canonical* evaluator for each task, so everyone reports comparable numbers. The
+**data** lives on Hugging Face; the **leaderboard** lives in the dataset card.
+
+> **Dataset & leaderboard:** 🤗 [pdokoupil/SM-RS](https://huggingface.co/datasets/pdokoupil/SM-RS)
+> **Cite BOTH** the SM-RS 2.0 (TORS'26) and SM-RS (SIGIR'24) papers — see [Citing](#citing).
+
+## Install
+
+```bash
+pip install sm-rs                 # core (numpy / pandas / scikit-learn)
+pip install "sm-rs[lstm]"         # + TensorFlow, for the optional LSTM baseline
+```
+
+## Quick start (Task 1: propensity estimation)
+
+```python
+import numpy as np
+from smrs.tasks import task1_propensity as t1
+
+# reproducible 80/20 split (seed 2024), reading a local SM-RS copy for now
+X_train, X_test, y_train, y_test = t1.split(data_dir="path/to/sm-rs")
+
+# ... train your estimator, produce an (N, 4) array of [rel, div, nov, exp] ...
+predictions = np.full((len(y_test), 4), 0.25)        # placeholder
+
+print(t1.evaluate(predictions, y_test))   # {'MAE': ..., 'MSE': ..., 'KLDiv': ...}
+```
+
+## The six tasks
+
+All six draw on the **same** dataset; they differ in what you predict and how it's
+scored. The canonical evaluator for each lives in `smrs.tasks.*`.
+
+| # | Task | You produce | Metric(s) | Evaluator |
+|---|------|-------------|-----------|-----------|
+| 1 | **Propensity estimation** | a 4-vector propensity per user | MAE · MSE · KL | ✅ `task1_propensity` |
+| 2 | **Results proportionality** | a top-k list matching target propensities | MAE · KL · wSUM · Pearson ρ | ✅ `task2_proportionality`¹ |
+| 3 | **Selections-aware reranking** | a reranked impression list | nDCG@10 · Precision@5 | ✅ `task3_reranking` |
+| 4 | **Diversity-metric definition** | per-list diversity values | MAE · MSE · KL | ✅ `task4_diversity` |
+| 5 | **Perceived quality** (5.1 rel / 5.2 div / 5.3 nov / 5.4 ser) | per-objective perception | MAE · MSE · Kendall τ | ✅ `task5_perceived` |
+| 6 | **Satisfaction** (6.1 / 6.2) | overall satisfaction | MAE · MSE · Kendall τ | ✅ `task6_satisfaction` |
+
+¹ Task 2's *metric layer* is implemented; turning a top-k list into achieved
+objective proportions needs the derived matrices (see Data) and lands with the
+source→rating-matrix builder.
+
+## Data
+
+Two layers:
+
+1. **Core tables** (the collected study data, **CC-BY**, hosted on Hugging Face):
+   `behaviors`, `propensities`, `objective_perceptions`, `criteria_values`,
+   `comparative_diversity`, `users`, `movies`, `books`. Items are referenced by ID.
+
+   **Auto-downloaded** from the Hub and cached — no manual download:
+
+   ```python
+   from smrs import data
+   df = data.load("propensities")                 # downloads from HF, cached
+   # offline / local copy (e.g. OSF download):
+   df = data.load("propensities", data_dir="path/to/sm-rs")   # or set $SMRS_DATA_DIR
+   ```
+
+   Users of the 🤗 `datasets` library can equivalently do
+   `load_dataset("pdokoupil/SM-RS", "behaviors")`.
+
+2. **Derived matrices** — *recomputed locally, not downloaded.* The list-scoring
+   tasks (2, 3) need per-item / per-pair artifacts (relevance via item-item,
+   intra-list diversity via a distance matrix, novelty via popularity). Rather than
+   ship multi-GB blobs — or redistribute the third-party catalogs they come from —
+   the benchmark **recomputes** them from a rating matrix you build from your own
+   download of the public source datasets (movies: **MovieLens 25M** — the "Latest"
+   snapshot at collection time — plus **MovieLens Tag Genome 2021**; books:
+   **goodbooks-10k**):
+
+   ```python
+   from smrs import derived
+   art = derived.build_artifacts(rating_matrix_movies)   # {item_item, distance_matrix, mean_popularities}
+   ```
+
+   `derived` provides the deterministic pieces: `popularity`, `cosine_distance`
+   (1 − cosine over item rating-vectors), and `ease_item_item` (EASE^R closed form,
+   used as the relevance model). This keeps the benchmark lightweight and
+   license-clean. The **bit-exact original artifacts are archived on OSF**
+   ([v2](https://osf.io/wsakx)) for strict reproduction.
+
+   Get the source datasets with the bundled fetcher (downloads from the official
+   hosts — GroupLens, the goodbooks repo — under their licenses; or place them
+   there manually):
+
+   ```bash
+   smrs-fetch --list                      # show sources + licenses, no download
+   smrs-fetch --dest ./sm-rs-sources      # download MovieLens 25M, Tag Genome 2021, goodbooks-10k
+   ```
+
+> **Why recompute?** MovieLens may not be redistributed, and a 5 GB download hurts
+> adoption. You obtain MovieLens/goodbooks under their own licenses; we ship only
+> the study data, the id-maps (`movies.json`: movieId→imdbId; `books.json`:
+> book_index→goodreads_id), and the recompute code. The canonical scorer is pinned
+> (sources above; **positive feedback = rating ≥ 3**; EASE λ) so results stay
+> comparable. For strict reproduction of the paper's numbers, use the OSF artifacts.
+
+## Reproduction check
+
+`examples/reproduce_perceived.py` reproduces the paper's **Linear Regression**
+baseline for Tasks 5 & 6 (the one baseline needing no derived/source data), scored
+with this package's evaluators — MAE matches the paper exactly:
+
+| subtask | MAE (ours / paper) | MSE | Kendall τ |
+|---|---|---|---|
+| 5.1 relevance | 0.235 / 0.235 | 0.086 / 0.085 | 0.076 / 0.080 |
+| 5.2 diversity | 0.222 / 0.222 | 0.080 / 0.061 | 0.197 / 0.196 |
+| 5.3 novelty | 0.259 / 0.259 | 0.104 / 0.104 | 0.143 / 0.143 |
+| 5.4 serendipity | 0.270 / 0.270 | 0.104 / 0.103 | 0.036 / 0.039 |
+| 6 satisfaction | 0.255 / 0.255 | 0.102 / 0.102 | 0.039 / 0.045 |
+
+```bash
+SMRS_DATA_DIR=/path/to/sm-rs python examples/reproduce_perceived.py
+```
+
+## Submitting to the leaderboard
+
+Self-service (no submission server): run the canonical `evaluate()` for a task,
+then open a PR adding your row to the leaderboard in the
+[dataset card](https://huggingface.co/datasets/pdokoupil/SM-RS), with a link to
+reproduce. The shipped baselines are the rows to beat.
+
+## Citing
+
+Please cite **both** papers (GitHub's "Cite this repository" reads
+[`CITATION.cff`](CITATION.cff)):
+
+```bibtex
+@article{dokoupil2026smrs2,
+  author  = {Dokoupil, Patrik and Peska, Ladislav},
+  title   = {SM-RS 2.0: User-perceived Qualities of Single- and Multi-Objective Recommender Systems},
+  journal = {ACM Transactions on Recommender Systems},
+  volume  = {4}, number = {3}, year = {2026},
+  doi     = {10.1145/3754459}
+}
+@inproceedings{dokoupil2024smrs,
+  author    = {Dokoupil, Patrik and Peska, Ladislav and Boratto, Ludovico},
+  title     = {SM-RS: Single- and Multi-Objective Recommendations with Contextual Impressions and Beyond-Accuracy Propensity Scores},
+  booktitle = {Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
+  series    = {SIGIR '24}, pages = {988--995}, year = {2024},
+  doi       = {10.1145/3626772.3657863}
+}
+```
+
+## License
+
+Code: MIT (see [`LICENSE`](LICENSE)). Data: CC-BY-4.0 (see the dataset card).