Add MAGE datasets with original features and interaction splits

README.md

@@ -0,0 +1,149 @@
+---
+pretty_name: MAGE multimodal recommendation datasets
+language:
+- en
+tags:
+- recommendation
+- multimodal
+- graph-learning
+- arxiv:2610.37409
+configs:
+- config_name: baby
+  default: true
+  data_files:
+  - split: train
+    path: data/baby/train.parquet
+  - split: validation
+    path: data/baby/validation.parquet
+  - split: test
+    path: data/baby/test.parquet
+- config_name: sports
+  data_files:
+  - split: train
+    path: data/sports/train.parquet
+  - split: validation
+    path: data/sports/validation.parquet
+  - split: test
+    path: data/sports/test.parquet
+- config_name: clothing
+  data_files:
+  - split: train
+    path: data/clothing/train.parquet
+  - split: validation
+    path: data/clothing/validation.parquet
+  - split: test
+    path: data/clothing/test.parquet
+- config_name: elec
+  data_files:
+  - split: train
+    path: data/elec/train.parquet
+  - split: validation
+    path: data/elec/validation.parquet
+  - split: test
+    path: data/elec/test.parquet
+---
+
+# MAGE datasets
+
+Prepared Baby, Sports, Clothing, and Electronics datasets for **Can Complementary Signals Bridge Similarity Islands? Manifold-Augmented Graph Embedding for Multimodal Recommendation**.
+
+- [Paper](https://arxiv.org/abs/2610.37409)
+- [Code and training instructions](https://github.com/Alex-bupt/mage)
+- [Input format](https://github.com/Alex-bupt/mage/blob/main/docs/DATA_FORMAT.md)
+
+The release contains fixed interaction splits, item and user ID mappings, visual and text feature matrices, an expanded text-feature catalog, and offline relation graphs. It provides numeric features rather than raw images or review text. These prepared Amazon-category benchmarks build on the data used by [GUME](https://github.com/Alex-bupt/mage/blob/main/docs/upstream/GUME_README.md) and MMRec.
+
+## Contents
+
+`archive/MAGE_datasets_2026-10-08.zip` is an unchanged mirror of the prepared dataset archive previously distributed through Google Drive. Extract it with a ZIP64-compatible utility to obtain:
+
+```text
+MAGE_datasets/
+  baby/
+  sports/
+  clothing/
+  elec/
+  README.txt
+  SHA256SUMS.txt
+```
+
+Electronics uses `elec` as its directory and dataset name in the code.
+
+| Dataset | Recommendation items | Expanded text rows | Stage-1 `--seed-max` |
+| --- | ---: | ---: | ---: |
+| Baby | 7,050 | 71,317 | 7,049 |
+| Sports | 18,357 | 77,832 | 18,356 |
+| Clothing | 23,033 | 84,695 | 23,032 |
+| Electronics | 63,001 | 498,196 | 63,000 |
+
+Each dataset directory includes:
+
+| File | Contents |
+| --- | --- |
+| `<dataset>.inter` | Original interaction table and split labels |
+| `<dataset>_hard_eval.inter` | Same training/validation rows, with a restricted hard test subset |
+| `i_id_mapping.csv`, `u_id_mapping.csv` | Original catalog/user identifiers and zero-based model IDs |
+| `image_feat.npy` | Aligned visual features, `(items, 4096)`, float64 |
+| `text_feat.npy` | Augmented text features, `(items, 512)`, float32 |
+| `text_feat_raw.npy` | Unmodified recommendation-item text features |
+| `text_feat_full.npy`, `i_id_mapping_full.csv` | Expanded text catalog and aligned item mapping |
+| `related_by_id.json` | Relations over the expanded item IDs |
+| `p2b_map_filtered.json` | Complementary edges between recommendation items |
+| `offline_manifest.json` | Preparation settings, shapes and feature hashes |
+
+The `.inter` files and the mapping `.csv` files are **tab-separated**. Interaction columns are `userID`, `itemID`, `rating`, `timestamp`, and `x_label`. The split labels are `0` = training, `1` = validation, and `2` = test. Keep the supplied IDs, row order and splits unchanged. The user mapping retains the original benchmark's pseudonymous user identifiers.
+
+## Explore interaction splits
+
+The Parquet files under `data/` expose the original interaction table as four Hugging Face configurations: `baby`, `sports`, `clothing`, and `elec`. Each retains the five source columns and separates rows according to the existing `x_label`; no split is resampled. These tables also power the dataset viewer.
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("Alex6657/MAGE", "baby")
+print(dataset["train"][0])
+```
+
+| Configuration | Users | Training rows | Validation rows | Test rows |
+| --- | ---: | ---: | ---: | ---: |
+| `baby` | 19,445 | 118,551 | 20,559 | 21,682 |
+| `sports` | 35,598 | 218,409 | 37,899 | 40,029 |
+| `clothing` | 39,387 | 197,338 | 40,150 | 41,189 |
+| `elec` | 192,403 | 1,304,382 | 192,403 | 192,403 |
+
+The Parquet configurations contain interactions only. Download the archive for feature matrices, mappings, graphs and the separate hard-test files needed for the full MAGE workflow.
+
+## Download the training data
+
+```python
+from huggingface_hub import hf_hub_download
+
+archive = hf_hub_download(
+    repo_id="Alex6657/MAGE",
+    repo_type="dataset",
+    filename="archive/MAGE_datasets_2026-10-08.zip",
+)
+print(archive)
+```
+
+The archive is **1,882,496,634 bytes**. SHA-256:
+
+```text
+8412c252d01efbcadc48101b3d3634191318e554d11219a93119b6253fb0651a
+```
+
+The extracted archive contains `SHA256SUMS.txt` for its files. From `MAGE_datasets/`, verify it with `sha256sum -c SHA256SUMS.txt` where available.
+
+## Prepare text features for training
+
+The archived `text_feat.npy` files use Stage-1 gap retention **0.60**. The current Baby, Sports and Clothing configurations in the code repository use **0.20**, **0.40** and **0.40**, respectively. For those configurations, regenerate `text_feat.npy` from `text_feat_full.npy` using the repository's Stage-1 commands before training. Preserve the original files and use a fresh working directory without graph caches.
+
+`--seed-max` is the inclusive last recommendation item ID. Expanded text rows supply neighbor features and must not be passed directly as the Stage-2 training feature matrix. The first recommendation-item rows of the expanded mapping must match the seed mapping. All text embeddings have 512 columns; visual features have 4096 columns.
+
+Electronics has 480 zero visual-feature rows, representing missing visual signal.
+
+## Use and attribution
+
+These files are intended for research on multimodal recommendation. Preserve the original benchmark attribution and applicable third-party terms. This mirror does not grant additional rights to the underlying datasets. The code repository retains its [upstream references and notices](https://github.com/Alex-bupt/mage/blob/main/docs/upstream/GUME_README.md).
+
+Please cite the [MAGE paper](https://arxiv.org/abs/2610.37409) when using the prepared features and graphs, together with the relevant original data and method sources.

archive/MAGE_datasets_2026-10-08.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8412c252d01efbcadc48101b3d3634191318e554d11219a93119b6253fb0651a
+size 1882496634

archive/MAGE_datasets_2026-10-08.zip.sha256

@@ -0,0 +1 @@
+8412c252d01efbcadc48101b3d3634191318e554d11219a93119b6253fb0651a  MAGE_datasets_2026-10-08.zip

archive/README.txt

@@ -0,0 +1,50 @@
+MAGE prepared datasets (2026-10-08)
+Source code: https://github.com/Alex-bupt/mage
+Validated code revision: 7434e84212155dc9b954a1ed171b65c8bfcc2d2d
+
+Extract the ZIP using a ZIP64-compatible utility. MAGE_datasets/ is the data
+root containing baby/, sports/, clothing/ and elec/. From the code repository:
+python main/src/main.py --model mage --dataset clothing --config main/src/configs/experiments/clothing_local.yaml --data-path /absolute/path/to/MAGE_datasets
+
+Each dataset includes:
+  <name>.inter: original training/validation/test split
+  <name>_hard_eval.inter: same training/validation rows, restricted hard test set
+  i_id_mapping.csv, u_id_mapping.csv: recommendation item and user mappings
+  image_feat.npy: aligned visual features (N x 4096, float64)
+  text_feat.npy: enhanced training text features (N x 512, float32)
+  text_feat_raw.npy: unmodified seed text features for baseline analysis
+  text_feat_full.npy: expanded text catalog, including offline neighbor items
+  i_id_mapping_full.csv: expanded catalog item mapping
+  related_by_id.json: expanded-ID offline relation graph
+  p2b_map_filtered.json: seed-ID online complementary graph
+  offline_manifest.json: algorithm parameters and feature provenance
+
+Dataset     Seed items  Expanded text rows  --seed-max
+baby             7050               71317        7049
+sports          18357               77832       18356
+clothing        23033               84695       23032
+elec            63001              498196       63000
+
+The training text features are already augmented. To regenerate Clothing:
+python stage1_process.py --emb-path /absolute/path/to/MAGE_datasets/clothing/text_feat_full.npy --related-path /absolute/path/to/MAGE_datasets/clothing/related_by_id.json --out-path /absolute/path/to/MAGE_datasets/clothing/text_feat.npy --seed-max 23032
+Use the matching last seed ID from the table for other datasets. Do not use
+expanded text features directly as the training feature matrix.
+
+Shared offline parameters: STRONG_PERCENTILE=95, WEAK_PERCENTILE=45,
+WEAK_SIM_MIN=0.20, CONCEPTOR_APERTURE=12, GAP_KEEP_RATIO=0.60,
+MARGIN_EPS=1e-8; fixed ALPHA_HARD_CAP=0.35. The code applies WEAK_SIM_MIN
+as a hard similarity floor; its inherited soft-weighting comment is stale.
+
+Validation: mappings, feature row alignment, finite values, split membership
+and hard-test subset consistency were checked. All four datasets completed
+one CPU epoch plus validation/test on real 32-user subsets with reduced model
+dimensions. This does not establish full-dataset convergence or paper metrics.
+Electronics contains 480 zero visual-feature rows; normalization guards handle
+these missing visual signals. No synthetic test data or model caches are included.
+
+SHA256SUMS.txt covers every other file in the extracted MAGE_datasets directory.
+Run sha256sum -c SHA256SUMS.txt from that directory where available.
+Keep IDs and row order unchanged. Clear/rebuild cached KNN graphs if features
+are replaced. See repository docs/DATA_FORMAT.md for loader details and
+docs/upstream/GUME_README.md for upstream attribution and academic-use notice.
+This package grants no additional rights to underlying third-party datasets.

data/baby/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:05c83d1a43c2a30b200768142446b0d13ac3d61abaa2b4856c4e19f1a31b51c8
+size 156107

data/baby/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b7f1c25afc9cfdabbcc98e1a98323aadafcb6b3a9fd72960f45864e94f4bb15
+size 571663

data/baby/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:197fe34fd11b459b7904ee3925fe3a2867b910deb656d8f0a27757e7eb829265
+size 149776

data/clothing/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7fbf81b4e9fc62ad5c041cff9cec9403f3b2e0d72ea2712eef3e123820f4b441
+size 309072

data/clothing/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1f62ecbd7fd730714c0f9752cae53df8ca43355bfda2e393901e2538189e6771
+size 864297

data/clothing/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed59ffed5ce5a5ad447ff7be2ad80d5d116b0b430ec0c4cd601b52c629fd3114
+size 303401

data/elec/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b72b8235cb430b79368551b3bd5f488bfbac6acb32a640847b5b256cf444a906
+size 1243759

data/elec/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29371ff3b6a3fd475c9709e2be0eba35daedee0ca846892fe9c8c10cdfdde718
+size 6306395

data/elec/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:86df33d305f87aac83ae2ea5506134ef1c5e797f37cc5e2a20557a65a7b1dde5
+size 1257543

data/sports/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f8438c202603cee709aec729ef3fa020d4cbdbd85e92b4728337d33daae08eb2
+size 288249

data/sports/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c290dc9e7c71665467f153392fd113f6cd41d6c6ebedc94c4e4446df647779db
+size 942004

data/sports/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2ce274a214d801d4c88d584d6f3a3e3d979c97130149a26724fe4de867c9ca1
+size 276030

dataset_statistics.json

@@ -0,0 +1,74 @@
+{
+  "baby": {
+    "rows": 160792,
+    "users": 19445,
+    "items": 7050,
+    "splits": {
+      "train": 118551,
+      "validation": 20559,
+      "test": 21682
+    },
+    "source_member": "MAGE_datasets/baby/baby.inter",
+    "columns": {
+      "userID": "int64",
+      "itemID": "int64",
+      "rating": "float64",
+      "timestamp": "int64",
+      "x_label": "int64"
+    }
+  },
+  "sports": {
+    "rows": 296337,
+    "users": 35598,
+    "items": 18357,
+    "splits": {
+      "train": 218409,
+      "validation": 37899,
+      "test": 40029
+    },
+    "source_member": "MAGE_datasets/sports/sports.inter",
+    "columns": {
+      "userID": "int64",
+      "itemID": "int64",
+      "rating": "float64",
+      "timestamp": "int64",
+      "x_label": "int64"
+    }
+  },
+  "clothing": {
+    "rows": 278677,
+    "users": 39387,
+    "items": 23033,
+    "splits": {
+      "train": 197338,
+      "validation": 40150,
+      "test": 41189
+    },
+    "source_member": "MAGE_datasets/clothing/clothing.inter",
+    "columns": {
+      "userID": "int64",
+      "itemID": "int64",
+      "rating": "float64",
+      "timestamp": "int64",
+      "x_label": "int64"
+    }
+  },
+  "elec": {
+    "rows": 1689188,
+    "users": 192403,
+    "items": 63001,
+    "splits": {
+      "train": 1304382,
+      "validation": 192403,
+      "test": 192403
+    },
+    "source_member": "MAGE_datasets/elec/elec.inter",
+    "columns": {
+      "userID": "int64",
+      "itemID": "int64",
+      "rating": "float64",
+      "timestamp": "int64",
+      "x_label": "int64"
+    }
+  }
+}