Datasets

Available loaders, metadata, and measured defaults. See Datasets for citations, Checkpoint CLI Reference for parameters, and Examples for code.

Dataset capabilities

Dataset

Item metadata

Precomputed embeddings importable here

Interaction timestamps

MovieLens 1M

Titles, genres, descriptions; optional features from plots/posters/trailers

Text, image, audio, video (SWAP)

Yes

MovieLens 20M

Titles, genres, descriptions; optional tag annotations

None integrated

Yes

Goodbooks-10k

Titles, authors, descriptions; optional tag annotations

None integrated

No

Amazon Reviews 2023

Product text and optional image URLs

Compute text/image embeddings yourself

Yes

Steam

Titles, genres, tags, developers and other game fields

None integrated

Yes, day precision

Netflix Prize

Movie titles and years

None integrated

Yes, day precision

MSD Taste Profile

Item IDs only

None integrated

No

Gowalla

Location coordinates

None integrated

Yes

DBbook

Book titles and DBpedia mappings; optional abstract/cover features

Text, image (SWAP)

No

Last.fm-2K

Artist names and page/image URLs; optional tag/cover/song features

Text, image, audio (SWAP)

No listening timestamps

Reading the default tables

  • Support: minimum user/item interactions. Totals are after preprocessing; temporal totals span all stages. DBbook official totals use training data only.

  • Val/Test: eligible users, distinct target items, and cold % (items absent from original training interactions). † means fewer than 1,000 users.

  • Metadata: union of measured catalogs, including preprocessed items. Image URLs are not validated; text means constructed entity_text. Feature coverage uses this union; d is vector dimension. Unknown is not zero.

  • Splits: random item defaults are 85/5/10 train/val/test. Random splits deduplicate repeated user/item pairs; ordered splits retain events.

Defaults binarize retained feedback; measured does not mean paper-reproducing. See Interpreting the documentation tables for definitions, methodology, and the non-Amazon measurement archive.

MovieLens 1M

Loading: dataset="ml1m" downloads GroupLens ratings and beeFormer descriptions; SWAP features are optional. A temporary expired-certificate workaround applies.

Terms: MovieLens research-use conditions; commercial use requires permission. Cite the description/feature sources too.

Defaults: Positive ratings and support filtering retain useful histories; the 30-word minimum selects descriptive metadata. The temporal default is poorly sized here: only 91 training users.

Installed MovieLens 1M preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
MovieLens 1M
Rating ≥4
Min text: 30 words
Seed: 42
5/1
6,033 users
3,295 items
Train: 4,533 users
Val: 500 users †
1,860 items / 0.00% cold
Test: 1,000 users
2,256 items / 0.00% cold
5/1
6,033 users
3,295 items
Train: 6,033 users
Val: 5,338 users
165 items / 100.00% cold
Test: 5,643 users
330 items / 100.00% cold
5/1
6,033 users
3,295 items
Train: 6,033 users
Val: 6,033 users
1,722 items / 0.29% cold
Test: 6,033 users
1,758 items / 0.17% cold
5/1
943 users
3,117 items
Train: 91 users
Val: 771 users †
2,439 items / 26.98% cold
Test: 435 users †
2,033 items / 19.43% cold
Window: 8136 h
Union: 3,295 items
Image URLs: not exposed
≥10 words: 3,295 (100.00%)
Precomputed features:
text/minilm: 3,295 (100.00%), 384d
text/mpnet: 3,295 (100.00%), 768d
image/resnet152: 2,942 (89.29%), 1,000d
image/vgg: 2,942 (89.29%), 4,096d
image/vit_cls: 2,942 (89.29%), 768d
image/vit_avg: 2,942 (89.29%), 768d
audio/vggish: 3,171 (96.24%), 128d
audio/whisper: 3,171 (96.24%), 512d
video/i3d: 3,170 (96.21%), 2,048d
video/r2p1d: 3,171 (96.24%), 512d

MovieLens 20M

Loading: dataset="ml20m" downloads GroupLens ratings and beeFormer descriptions; user-tag annotations are optional. The same expired-certificate workaround applies.

Terms: GroupLens usage conditions; commercial use requires permission. Cite beeFormer when using its descriptions.

Defaults: MultVAE-style positive-feedback filtering, plus a 30-word text minimum. Holdout sizes are practical defaults, not the paper’s exact split.

Installed MovieLens 20M preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
MovieLens 20M
Rating ≥4
Min text: 30 words
Seed: 42
5/1
136,589 users
16,902 items
Train: 129,089 users
Val: 2,500 users
4,549 items / 0.00% cold
Test: 5,000 users
5,972 items / 0.00% cold
5/1
136,589 users
16,902 items
Train: 136,589 users
Val: 102,285 users
846 items / 100.00% cold
Test: 124,822 users
1,691 items / 100.00% cold
5/1
136,589 users
16,902 items
Train: 136,589 users
Val: 136,589 users
7,353 items / 0.60% cold
Test: 136,589 users
7,784 items / 0.90% cold
5/1
5,903 users
15,399 items
Train: 3,801 users
Val: 3,323 users
6,705 items / 12.17% cold
Test: 3,217 users
7,346 items / 18.46% cold
Window: 8136 h
Union: 16,902 items
Image URLs: not exposed
≥10 words: 16,902 (100.00%)

Goodbooks-10k

Loading: dataset="goodbooks" downloads book ratings/metadata and beeFormer descriptions; tag annotations are optional.

Terms: CC BY-SA 4.0 for the upstream release; cite the additional description source.

Defaults: Positive ratings, minimum interaction support, and 30-word metadata give a text-ready benchmark. No timestamps: user/item splits only.

Installed Goodbooks-10k preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
Goodbooks-10k
Rating ≥4
Min text: 30 words
Seed: 0
5/1
53,365 users
9,975 items
Train: 49,865 users
Val: 1,000 users
5,361 items / 0.00% cold
Test: 2,500 users
7,820 items / 0.00% cold
5/1
53,365 users
9,975 items
Train: 53,365 users
Val: 51,301 users
499 items / 100.00% cold
Test: 53,070 users
998 items / 100.00% cold
Not supported
No interaction timestamps
Not supported
No interaction timestamps
Union: 9,975 items
Image URLs: not exposed
≥10 words: 9,975 (100.00%)

Amazon Reviews 2023

Loading: dataset="amazon2023" downloads one category’s ratings and item metadata, not review text. include_image_urls=True retains image links.

Terms: Follow the Amazon Reviews 2023 release’s usage conditions and source attribution.

Defaults: All ratings count as feedback. Category-specific support filters keep graphs manageable while preserving evaluation users; no sampling or text minimum. Text fields also vary by category. Small holdouts remain flagged (†).

Details: Amazon default-setting audit, Amazon text and image coverage, and measured profiles.

Installed category-specific Amazon support defaults

Category

User split

Item split

Leave-last-out

Temporal

Item metadata (union of splits)

All_Beauty
2/2
22,634 users
12,105 items
Val: 2,097 users
1,826 items / 0.00% cold
Test: 2,068 users
1,854 items / 0.00% cold
2/2
22,634 users
12,105 items
Val: 2,414 users
602 items / 100.00% cold
Test: 4,813 users
1,203 items / 100.00% cold
4/1
2,909 users
10,266 items
Val: 2,909 users
2,503 items / 60.53% cold
Test: 2,909 users
2,487 items / 64.41% cold
3/1
3,622 users
10,734 items
Val: 1,410 users
1,657 items / 81.11% cold
Test: 670 users †
618 items / 89.97% cold
Union: 18,755 items
≥1 image: 18,755 (100.00%)
≥10 words: 18,528 (98.79%)

Amazon_Fashion
2/4
35,714 users
10,217 items
Val: 3,551 users
2,779 items / 0.00% cold
Test: 3,560 users
2,692 items / 0.00% cold
2/4
35,714 users
10,217 items
Val: 3,762 users
511 items / 100.00% cold
Test: 6,772 users
1,022 items / 100.00% cold
4/2
1,717 users
4,098 items
Val: 1,717 users
1,292 items / 15.48% cold
Test: 1,717 users
1,190 items / 21.18% cold
2/2
3,157 users
4,080 items
Val: 1,116 users
673 items / 72.07% cold
Test: 791 users †
467 items / 80.09% cold
Union: 13,201 items
≥1 image: 13,201 (100.00%)
≥10 words: 13,093 (99.18%)

Appliances
3/2
52,101 users
19,327 items
Val: 4,872 users
3,391 items / 0.00% cold
Test: 4,859 users
3,277 items / 0.00% cold
3/2
52,101 users
19,327 items
Val: 9,980 users
967 items / 100.00% cold
Test: 16,170 users
1,933 items / 100.00% cold
4/2
14,826 users
10,896 items
Val: 14,826 users
6,274 items / 14.87% cold
Test: 14,826 users
6,215 items / 17.28% cold
2/2
71,639 users
16,820 items
Val: 28,766 users
8,307 items / 37.52% cold
Test: 20,484 users
7,199 items / 47.84% cold
Union: 22,334 items
≥1 image: 22,334 (100.00%)
≥10 words: 22,323 (99.95%)

Arts_Crafts_and_Sewing
5/16
97,533 users
18,043 items
Val: 5,000 users
6,506 items / 0.00% cold
Test: 5,000 users
6,613 items / 0.00% cold
5/16
97,533 users
18,043 items
Val: 31,814 users
903 items / 100.00% cold
Test: 52,915 users
1,805 items / 100.00% cold
5/16
97,533 users
18,043 items
Val: 97,533 users
16,863 items / 0.00% cold
Test: 97,533 users
16,459 items / 0.00% cold
3/12
96,749 users
12,713 items
Val: 42,979 users
9,307 items / 21.85% cold
Test: 29,693 users
9,221 items / 37.64% cold
Union: 19,690 items
≥1 image: 19,680 (99.95%)
≥10 words: 19,688 (99.99%)

Automotive
7/25
99,264 users
16,079 items
Val: 5,000 users
7,492 items / 0.00% cold
Test: 5,000 users
7,574 items / 0.00% cold
7/25
99,264 users
16,079 items
Val: 42,440 users
804 items / 100.00% cold
Test: 66,549 users
1,608 items / 100.00% cold
7/25
99,264 users
16,079 items
Val: 99,264 users
15,101 items / 0.00% cold
Test: 99,264 users
14,639 items / 0.00% cold
6/12
86,556 users
19,182 items
Val: 49,082 users
15,181 items / 25.95% cold
Test: 30,194 users
12,891 items / 29.71% cold
Union: 20,809 items
≥1 image: 20,808 (>99.99%)
≥10 words: 20,808 (>99.99%)

Baby_Products
6/9
85,240 users
17,987 items
Val: 5,000 users
6,029 items / 0.00% cold
Test: 5,000 users
6,056 items / 0.00% cold
6/9
85,240 users
17,987 items
Val: 29,585 users
900 items / 100.00% cold
Test: 49,964 users
1,799 items / 100.00% cold
6/9
85,240 users
17,987 items
Val: 85,240 users
14,115 items / 0.03% cold
Test: 85,240 users
13,560 items / 0.03% cold
3/7
95,735 users
13,365 items
Val: 38,591 users
7,822 items / 22.00% cold
Test: 25,313 users
8,198 items / 43.40% cold
Union: 20,108 items
≥1 image: 20,105 (99.99%)
≥10 words: 20,103 (99.98%)

Beauty_and_Personal_Care
8/25
95,840 users
16,817 items
Val: 5,000 users
7,981 items / 0.00% cold
Test: 5,000 users
8,008 items / 0.00% cold
8/25
95,840 users
16,817 items
Val: 44,765 users
841 items / 100.00% cold
Test: 66,388 users
1,682 items / 100.00% cold
8/25
95,840 users
16,817 items
Val: 95,840 users
15,180 items / 0.00% cold
Test: 95,840 users
14,750 items / 0.00% cold
6/18
95,044 users
14,656 items
Val: 50,440 users
9,685 items / 28.90% cold
Test: 40,215 users
11,789 items / 48.75% cold
Union: 19,153 items
≥1 image: 19,153 (100.00%)
≥10 words: 19,153 (100.00%)

Books
5/17
428,352 users
94,864 items
Val: 20,000 users
31,567 items / 0.00% cold
Test: 20,000 users
32,453 items / 0.00% cold
5/17
428,352 users
94,864 items
Val: 161,770 users
4,744 items / 100.00% cold
Test: 241,076 users
9,487 items / 100.00% cold
5/17
428,352 users
94,864 items
Val: 428,352 users
80,327 items / 0.04% cold
Test: 428,352 users
76,778 items / 0.05% cold
6/6
112,433 users
96,465 items
Val: 63,581 users
37,923 items / 28.87% cold
Test: 41,812 users
26,194 items / 40.21% cold
Union: 127,961 items
≥1 image: 113,452 (88.66%)
≥10 words: 127,913 (99.96%)

CDs_and_Vinyl
5/16
66,625 users
18,586 items
Val: 5,000 users
8,295 items / 0.00% cold
Test: 5,000 users
8,493 items / 0.00% cold
5/16
66,625 users
18,586 items
Val: 24,546 users
930 items / 100.00% cold
Test: 38,473 users
1,859 items / 100.00% cold
5/16
66,625 users
18,586 items
Val: 66,625 users
16,271 items / <0.01% cold
Test: 66,625 users
16,039 items / 0.01% cold
2/5
29,237 users
13,650 items
Val: 13,120 users
6,986 items / 22.70% cold
Test: 6,481 users
4,566 items / 24.11% cold
Union: 23,072 items
≥1 image: 23,054 (99.92%)
≥10 words: 23,072 (100.00%)

Cell_Phones_and_Accessories
6/13
92,781 users
19,322 items
Val: 5,000 users
6,692 items / 0.00% cold
Test: 5,000 users
6,736 items / 0.00% cold
6/13
92,781 users
19,322 items
Val: 28,616 users
967 items / 100.00% cold
Test: 52,067 users
1,933 items / 100.00% cold
6/13
92,781 users
19,322 items
Val: 92,781 users
16,504 items / 0.04% cold
Test: 92,781 users
15,069 items / 0.05% cold
5/8
89,713 users
19,096 items
Val: 41,928 users
7,932 items / 38.87% cold
Test: 35,885 users
10,338 items / 66.97% cold
Union: 23,458 items
≥1 image: 23,458 (100.00%)
≥10 words: 23,454 (99.98%)

Clothing_Shoes_and_Jewelry
10/35
93,809 users
14,289 items
Val: 5,000 users
8,145 items / 0.00% cold
Test: 5,000 users
8,166 items / 0.00% cold
10/35
93,809 users
14,289 items
Val: 46,057 users
715 items / 100.00% cold
Test: 70,344 users
1,429 items / 100.00% cold
10/35
93,809 users
14,289 items
Val: 93,809 users
13,220 items / 0.00% cold
Test: 93,809 users
12,877 items / 0.00% cold
10/17
113,384 users
24,824 items
Val: 76,251 users
20,663 items / 54.84% cold
Test: 59,629 users
19,421 items / 57.30% cold
Union: 24,842 items
≥1 image: 24,839 (99.99%)
≥10 words: 24,842 (100.00%)

Digital_Music
2/2
2,782 users
2,389 items
Val: 254 users †
252 items / 0.00% cold
Test: 244 users †
240 items / 0.00% cold
2/2
2,782 users
2,389 items
Val: 314 users †
119 items / 100.00% cold
Test: 563 users †
236 items / 100.00% cold
4/1
2,243 users
16,240 items
Val: 2,243 users
2,184 items / 85.21% cold
Test: 2,243 users
2,177 items / 86.08% cold
2/1
2,115 users
8,154 items
Val: 862 users †
1,121 items / 92.51% cold
Test: 604 users †
735 items / 93.61% cold
Union: 18,995 items
≥1 image: 18,995 (100.00%)
≥10 words: 16,956 (89.27%)

Electronics
8/16
494,089 users
92,180 items
Val: 20,000 users
31,288 items / 0.00% cold
Test: 20,000 users
31,489 items / 0.00% cold
8/16
494,089 users
92,180 items
Val: 224,865 users
4,609 items / 100.00% cold
Test: 347,160 users
9,218 items / 100.00% cold
8/16
494,089 users
92,180 items
Val: 494,089 users
72,119 items / <0.01% cold
Test: 494,089 users
66,509 items / <0.01% cold
6/12
484,391 users
78,239 items
Val: 252,303 users
39,800 items / 21.04% cold
Test: 177,638 users
40,647 items / 40.37% cold
Union: 99,261 items
≥1 image: 99,256 (>99.99%)
≥10 words: 99,242 (99.98%)

Gift_Cards
2/1
11,426 users
749 items
Val: 1,135 users
230 items / 0.00% cold
Test: 1,131 users
240 items / 0.00% cold
2/2
11,333 users
575 items
Val: 2,285 users
29 items / 100.00% cold
Test: 1,445 users
58 items / 100.00% cold
4/1
1,201 users
503 items
Val: 1,201 users
253 items / 14.62% cold
Test: 1,201 users
244 items / 14.75% cold
2/1
2,313 users
421 items
Val: 1,092 users
123 items / 28.46% cold
Test: 577 users †
120 items / 38.33% cold
Union: 749 items
≥1 image: 749 (100.00%)
≥10 words: 748 (99.87%)

Grocery_and_Gourmet_Food
7/24
98,300 users
16,735 items
Val: 5,000 users
7,768 items / 0.00% cold
Test: 5,000 users
7,813 items / 0.00% cold
7/24
98,300 users
16,735 items
Val: 42,173 users
837 items / 100.00% cold
Test: 63,925 users
1,674 items / 100.00% cold
7/24
98,300 users
16,735 items
Val: 98,300 users
15,347 items / 0.00% cold
Test: 98,300 users
14,851 items / 0.00% cold
6/13
86,148 users
19,238 items
Val: 50,360 users
14,478 items / 32.93% cold
Test: 34,731 users
13,429 items / 39.73% cold
Union: 21,273 items
≥1 image: 21,262 (99.95%)
≥10 words: 21,266 (99.97%)

Handmade_Products
2/2
22,156 users
11,878 items
Val: 2,006 users
1,725 items / 0.00% cold
Test: 2,040 users
1,788 items / 0.00% cold
2/2
22,156 users
11,878 items
Val: 2,335 users
587 items / 100.00% cold
Test: 4,283 users
1,180 items / 100.00% cold
4/1
3,716 users
14,723 items
Val: 3,716 users
3,350 items / 76.24% cold
Test: 3,716 users
3,347 items / 77.14% cold
3/1
3,774 users
11,710 items
Val: 1,410 users
2,029 items / 88.66% cold
Test: 1,251 users
1,708 items / 92.62% cold
Union: 23,704 items
≥1 image: 23,704 (100.00%)
≥10 words: 23,703 (>99.99%)

Health_and_Household
10/20
84,204 users
19,922 items
Val: 5,000 users
9,180 items / 0.00% cold
Test: 5,000 users
8,986 items / 0.00% cold
10/20
84,204 users
19,922 items
Val: 44,334 users
997 items / 100.00% cold
Test: 63,720 users
1,993 items / 100.00% cold
10/20
84,204 users
19,922 items
Val: 84,204 users
16,439 items / 0.00% cold
Test: 84,204 users
15,703 items / 0.00% cold
8/15
88,598 users
19,907 items
Val: 50,052 users
13,166 items / 26.50% cold
Test: 38,620 users
14,545 items / 44.23% cold
Union: 23,856 items
≥1 image: 23,856 (100.00%)
≥10 words: 23,856 (100.00%)

Health_and_Personal_Care
2/1
18,396 users
17,288 items
Val: 1,041 users
940 items / 0.00% cold
Test: 1,038 users
877 items / 0.00% cold
2/1
18,396 users
17,288 items
Val: 1,758 users
784 items / 100.00% cold
Test: 3,890 users
1,585 items / 100.00% cold
4/1
1,109 users
3,448 items
Val: 1,109 users
911 items / 42.26% cold
Test: 1,109 users
857 items / 47.37% cold
2/1
5,291 users
7,604 items
Val: 2,100 users
1,603 items / 72.99% cold
Test: 1,040 users
762 items / 77.30% cold
Union: 17,288 items
≥1 image: 17,288 (100.00%)
≥10 words: 16,849 (97.46%)

Home_and_Kitchen
12/35
96,909 users
17,431 items
Val: 5,000 users
9,721 items / 0.00% cold
Test: 5,000 users
9,750 items / 0.00% cold
12/35
96,909 users
17,431 items
Val: 51,957 users
872 items / 100.00% cold
Test: 78,055 users
1,744 items / 100.00% cold
12/35
96,909 users
17,431 items
Val: 96,909 users
15,181 items / 0.00% cold
Test: 96,909 users
14,374 items / 0.00% cold
10/25
93,903 users
14,879 items
Val: 53,997 users
11,264 items / 20.23% cold
Test: 38,614 users
10,726 items / 28.82% cold
Union: 18,990 items
≥1 image: 18,990 (100.00%)
≥10 words: 18,990 (100.00%)

Industrial_and_Scientific
4/8
74,487 users
19,251 items
Val: 5,000 users
5,348 items / 0.00% cold
Test: 5,000 users
5,350 items / 0.00% cold
4/8
74,487 users
19,251 items
Val: 17,942 users
963 items / 100.00% cold
Test: 33,333 users
1,926 items / 100.00% cold
4/8
74,487 users
19,251 items
Val: 74,487 users
16,742 items / 0.12% cold
Test: 74,487 users
16,000 items / 0.14% cold
3/5
66,901 users
17,690 items
Val: 31,809 users
10,279 items / 36.56% cold
Test: 23,291 users
10,034 items / 52.33% cold
Union: 22,281 items
≥1 image: 22,281 (100.00%)
≥10 words: 22,278 (99.99%)

Kindle_Store
12/76
98,841 users
17,800 items
Val: 5,000 users
13,209 items / 0.00% cold
Test: 5,000 users
13,406 items / 0.00% cold
12/76
98,841 users
17,800 items
Val: 63,226 users
890 items / 100.00% cold
Test: 85,505 users
1,780 items / 100.00% cold
12/76
98,841 users
17,800 items
Val: 98,841 users
15,961 items / 0.00% cold
Test: 98,841 users
15,154 items / 0.00% cold
8/50
96,853 users
19,957 items
Val: 60,020 users
15,954 items / 16.75% cold
Test: 42,975 users
14,155 items / 24.56% cold
Union: 23,549 items
≥1 image: 21,033 (89.32%)
≥10 words: 23,549 (100.00%)

Magazine_Subscriptions
2/1
6,808 users
2,003 items
Val: 1,115 users
472 items / 0.00% cold
Test: 1,093 users
463 items / 0.00% cold
2/2
6,408 users
1,241 items
Val: 1,070 users
63 items / 100.00% cold
Test: 1,594 users
124 items / 100.00% cold
4/1
797 users
1,110 items
Val: 797 users †
353 items / 25.21% cold
Test: 797 users †
381 items / 30.45% cold
2/1
458 users
522 items
Val: 151 users †
117 items / 45.30% cold
Test: 74 users †
66 items / 33.33% cold
Union: 2,003 items
≥1 image: 2,003 (100.00%)
≥10 words: 1,870 (93.36%)

Movies_and_TV
10/36
97,871 users
18,324 items
Val: 5,000 users
10,254 items / 0.00% cold
Test: 5,000 users
10,070 items / 0.00% cold
10/36
97,871 users
18,324 items
Val: 50,964 users
917 items / 100.00% cold
Test: 76,392 users
1,833 items / 100.00% cold
10/36
97,871 users
18,324 items
Val: 97,871 users
16,454 items / 0.00% cold
Test: 97,871 users
15,927 items / 0.00% cold
5/12
60,122 users
17,522 items
Val: 31,406 users
10,612 items / 18.37% cold
Test: 16,203 users
6,941 items / 23.51% cold
Union: 24,791 items
≥1 image: 24,791 (100.00%)
≥10 words: 24,636 (99.37%)

Musical_Instruments
5/7
50,431 users
16,766 items
Val: 5,000 users
5,933 items / 0.00% cold
Test: 5,000 users
5,863 items / 0.00% cold
5/7
50,431 users
16,766 items
Val: 18,353 users
839 items / 100.00% cold
Test: 27,033 users
1,677 items / 100.00% cold
5/7
50,431 users
16,766 items
Val: 50,431 users
12,985 items / 0.05% cold
Test: 50,431 users
12,681 items / 0.05% cold
2/5
77,684 users
15,914 items
Val: 32,616 users
10,123 items / 23.50% cold
Test: 23,272 users
9,093 items / 33.53% cold
Union: 20,065 items
≥1 image: 20,061 (99.98%)
≥10 words: 20,057 (99.96%)

Office_Products
5/18
98,570 users
12,740 items
Val: 5,000 users
5,177 items / 0.00% cold
Test: 5,000 users
5,178 items / 0.00% cold
5/18
98,570 users
12,740 items
Val: 30,907 users
637 items / 100.00% cold
Test: 49,343 users
1,274 items / 100.00% cold
5/18
98,570 users
12,740 items
Val: 98,570 users
12,194 items / 0.02% cold
Test: 98,570 users
11,905 items / 0.02% cold
4/8
90,062 users
19,965 items
Val: 39,715 users
11,142 items / 22.83% cold
Test: 29,909 users
12,535 items / 47.53% cold
Union: 22,096 items
≥1 image: 22,029 (99.70%)
≥10 words: 22,095 (>99.99%)

Patio_Lawn_and_Garden
6/21
98,613 users
14,019 items
Val: 5,000 users
6,556 items / 0.00% cold
Test: 5,000 users
6,539 items / 0.00% cold
6/21
98,613 users
14,019 items
Val: 33,334 users
701 items / 100.00% cold
Test: 55,849 users
1,402 items / 100.00% cold
6/21
98,613 users
14,019 items
Val: 98,613 users
13,333 items / 0.00% cold
Test: 98,613 users
12,826 items / 0.00% cold
5/13
89,415 users
14,329 items
Val: 52,279 users
10,667 items / 35.60% cold
Test: 28,043 users
9,718 items / 46.72% cold
Union: 17,357 items
≥1 image: 17,352 (99.97%)
≥10 words: 17,357 (100.00%)

Pet_Supplies
10/16
89,150 users
19,880 items
Val: 5,000 users
8,436 items / 0.00% cold
Test: 5,000 users
8,533 items / 0.00% cold
10/16
89,150 users
19,880 items
Val: 46,784 users
994 items / 100.00% cold
Test: 68,722 users
1,988 items / 100.00% cold
10/16
89,150 users
19,880 items
Val: 89,150 users
16,106 items / 0.00% cold
Test: 89,150 users
15,361 items / 0.00% cold
8/12
88,279 users
18,400 items
Val: 47,781 users
11,921 items / 19.44% cold
Test: 34,883 users
12,385 items / 36.15% cold
Union: 22,828 items
≥1 image: 22,824 (99.98%)
≥10 words: 22,825 (99.99%)

Software
6/6
98,634 users
13,664 items
Val: 5,000 users
4,254 items / 0.00% cold
Test: 5,000 users
4,129 items / 0.00% cold
6/6
98,634 users
13,664 items
Val: 34,962 users
684 items / 100.00% cold
Test: 62,676 users
1,367 items / 100.00% cold
6/6
98,634 users
13,664 items
Val: 98,634 users
9,800 items / 0.17% cold
Test: 98,634 users
9,400 items / 0.27% cold
2/2
76,101 users
14,548 items
Val: 29,981 users
5,902 items / 22.89% cold
Test: 14,134 users
3,762 items / 24.83% cold
Union: 18,322 items
≥1 image: 18,321 (>99.99%)
≥10 words: 18,318 (99.98%)

Sports_and_Outdoors
6/17
99,084 users
19,023 items
Val: 5,000 users
7,398 items / 0.00% cold
Test: 5,000 users
7,431 items / 0.00% cold
6/17
99,084 users
19,023 items
Val: 35,955 users
952 items / 100.00% cold
Test: 55,955 users
1,903 items / 100.00% cold
6/17
99,084 users
19,023 items
Val: 99,084 users
17,601 items / 0.00% cold
Test: 99,084 users
16,816 items / 0.00% cold
5/9
61,892 users
18,592 items
Val: 29,250 users
10,372 items / 24.21% cold
Test: 20,849 users
11,653 items / 48.85% cold
Union: 25,641 items
≥1 image: 25,641 (100.00%)
≥10 words: 25,637 (99.98%)

Subscription_Boxes
2/1
572 users
327 items
Val: 42 users †
28 items / 0.00% cold
Test: 47 users †
36 items / 0.00% cold
2/1
572 users
327 items
Val: 56 users †
16 items / 100.00% cold
Test: 141 users †
31 items / 100.00% cold
4/1
24 users
86 items
Val: 24 users †
23 items / 78.26% cold
Test: 24 users †
21 items / 95.24% cold
2/1
152 users
191 items
Val: 59 users †
62 items / 77.42% cold
Test: 42 users †
34 items / 88.24% cold
Union: 327 items
≥1 image: 327 (100.00%)
≥10 words: 326 (99.69%)

Tools_and_Home_Improvement
8/26
94,834 users
17,307 items
Val: 5,000 users
8,377 items / 0.00% cold
Test: 5,000 users
8,199 items / 0.00% cold
8/26
94,834 users
17,307 items
Val: 45,023 users
866 items / 100.00% cold
Test: 66,710 users
1,731 items / 100.00% cold
8/26
94,834 users
17,307 items
Val: 94,834 users
15,754 items / 0.00% cold
Test: 94,834 users
15,151 items / 0.00% cold
6/18
99,815 users
15,514 items
Val: 52,511 users
11,518 items / 21.32% cold
Test: 37,242 users
11,481 items / 33.10% cold
Union: 19,936 items
≥1 image: 19,935 (>99.99%)
≥10 words: 19,936 (100.00%)

Toys_and_Games
6/22
98,065 users
15,609 items
Val: 5,000 users
7,090 items / 0.00% cold
Test: 5,000 users
7,031 items / 0.00% cold
6/22
98,065 users
15,609 items
Val: 34,026 users
781 items / 100.00% cold
Test: 56,767 users
1,561 items / 100.00% cold
6/22
98,065 users
15,609 items
Val: 98,065 users
14,818 items / <0.01% cold
Test: 98,065 users
14,342 items / <0.01% cold
5/10
55,927 users
17,369 items
Val: 23,364 users
8,553 items / 20.29% cold
Test: 20,412 users
11,428 items / 54.60% cold
Union: 23,646 items
≥1 image: 23,644 (>99.99%)
≥10 words: 23,644 (>99.99%)

Video_Games
5/7
88,081 users
19,227 items
Val: 5,000 users
5,574 items / 0.00% cold
Test: 5,000 users
5,551 items / 0.00% cold
5/7
88,081 users
19,227 items
Val: 30,322 users
962 items / 100.00% cold
Test: 48,790 users
1,923 items / 100.00% cold
5/7
88,081 users
19,227 items
Val: 88,081 users
15,526 items / 0.15% cold
Test: 88,081 users
15,042 items / 0.17% cold
3/4
70,806 users
16,463 items
Val: 30,103 users
7,650 items / 29.50% cold
Test: 21,917 users
7,253 items / 48.01% cold
Union: 23,083 items
≥1 image: 23,082 (>99.99%)
≥10 words: 23,057 (99.89%)

Steam

Loading: dataset="steam" downloads McAuley’s review and game-metadata archives, then caches parsed interactions as Parquet.

Terms: Retain upstream usage and citation information.

Defaults: Every review, including negative ones, counts as feedback (BERT4Rec’s convention). Support filtering removes short histories; no text minimum, so missing metadata does not erase interactions.

Installed Steam preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
Steam
All feedback
Min text: 0 words
Seed: 42
5/1
281,645 users
15,051 items
Train: 261,645 users
Val: 9,994 users
4,949 items / 0.00% cold
Test: 9,998 users
4,897 items / 0.00% cold
5/1
281,645 users
15,051 items
Train: 281,639 users
Val: 98,181 users
753 items / 100.00% cold
Test: 162,161 users
1,506 items / 100.00% cold
5/1
334,730 users
15,068 items
Train: 334,730 users
Val: 334,730 users
10,453 items / 2.06% cold
Test: 334,730 users
10,621 items / 3.29% cold
5/1
250,498 users
14,965 items
Train: 91,041 users
Val: 138,434 users
9,904 items / 35.00% cold
Test: 192,164 users
13,781 items / 54.84% cold
Window: 8136 h
Union: 15,068 items
Image URLs: not exposed
≥10 words: 9,727 (64.55%)

Netflix Prize

Loading: dataset="netflix" reads the original-format ratings archive from Internet Archive, including dates and movie titles/years.

Terms: Original Netflix Prize conditions apply; the archive mirror does not replace those terms.

Defaults: MultVAE-style positive-rating and support filtering. No text minimum: short movie titles remain usable metadata.

Installed Netflix Prize preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
Netflix Prize
Rating ≥4
Min text: 0 words
Seed: 98765
5/1
463,435 users
17,769 items
Train: 383,435 users
Val: 40,000 users
14,546 items / 0.00% cold
Test: 40,000 users
14,474 items / 0.00% cold
5/1
463,435 users
17,769 items
Train: 463,434 users
Val: 377,662 users
889 items / 100.00% cold
Test: 440,315 users
1,777 items / 100.00% cold
5/1
463,435 users
17,769 items
Train: 463,435 users
Val: 463,435 users
13,185 items / 0.00% cold
Test: 463,435 users
13,433 items / 0.00% cold
5/1
262,636 users
17,766 items
Train: 58,419 users
Val: 130,014 users
15,317 items / 22.21% cold
Test: 241,687 users
17,612 items / 31.77% cold
Window: 8136 h
Union: 17,769 items
Image URLs: not exposed
≥10 words: 581 (3.27%)

MSD Taste Profile

Loading: dataset="taste-profile" downloads the original user/song play-count ZIP, not the full MSD audio features or metadata.

Terms: MSD / Echo Nest usage conditions apply.

Defaults: MultVAE-style support (20 interactions per user, 200 per item) removes sparse tails; play counts become binary feedback. No timestamps or text.

Installed MSD Taste Profile preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
MSD Taste Profile
All feedback
Min text: 0 words
Seed: 98765
20/200
558,350 users
36,385 items
Train: 458,350 users
Val: 50,000 users
36,262 items / 0.00% cold
Test: 50,000 users
36,256 items / 0.00% cold
20/200
558,350 users
36,385 items
Train: 558,350 users
Val: 478,324 users
1,820 items / 100.00% cold
Test: 543,782 users
3,639 items / 100.00% cold
Not supported
No interaction timestamps
Not supported
No interaction timestamps
Union: 36,385 items
Image URLs: not exposed
≥10 words: 0 (0.00%)

Gowalla

Loading: dataset="gowalla" downloads raw SNAP check-ins and coordinates, not the social graph or pre-made LightGCN splits.

Terms: Retain SNAP’s source citation and usage information.

Defaults: NGCF-style 10/10 support; 720-hour (30-day) temporal windows fit the available check-in history.

Installed Gowalla preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
Gowalla
All feedback
Min text: 0 words
Seed: 42
10/10
29,858 users
40,988 items
Train: 9,858 users
Val: 9,966 users
28,823 items / 0.00% cold
Test: 9,961 users
28,653 items / 0.00% cold
10/10
29,858 users
40,988 items
Train: 29,858 users
Val: 19,376 users
2,050 items / 100.00% cold
Test: 25,848 users
4,099 items / 100.00% cold
10/10
52,985 users
121,866 items
Train: 52,985 users
Val: 52,985 users
36,247 items / 0.00% cold
Test: 52,985 users
36,292 items / 0.00% cold
10/10
16,862 users
21,503 items
Train: 8,533 users
Val: 10,557 users
14,713 items / 24.71% cold
Test: 13,091 users
18,537 items / 40.63% cold
Window: 720 h
Union: 121,866 items
Image URLs: not exposed
≥10 words: 0 (0.00%)

DBbook

Loading: dataset="dbbook" downloads SWAP’s reconstructed interaction ZIP and DBpedia title mappings; abstract/cover embeddings are optional.

Terms: Consult the SWAP release; original DBbook/DBpedia rights still apply.

Defaults: Keep positive feedback and sufficiently supported histories, without filtering short titles. official preserves the supplied test boundary; see DBbook official protocol for validation details.

Installed DBbook preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Official

Item metadata (union of measured splits)
DBbook
Rating ≥1
Min text: 0 words
Seed: 42
5/1
5,884 users
6,538 items
Train: 4,384 users
Val: 496 users †
916 items / 0.00% cold
Test: 998 users †
1,443 items / 0.00% cold
5/1
5,884 users
6,538 items
Train: 5,883 users
Val: 2,498 users
327 items / 100.00% cold
Test: 3,974 users
654 items / 100.00% cold
Not supported
No interaction timestamps
Not supported
No interaction timestamps
5/1
2,799 users
4,548 items
Train: 2,799 users
Val: 2,799 users
2,335 items / 20.00% cold
Test: 2,144 users
2,742 items / 6.78% cold
Users/items above: training-only preprocessing
Union: 6,538 items
Image URLs: not exposed
≥10 words: 64 (0.98%)
Precomputed features:
text/minilm: 5,698 (87.15%), 384d
text/mpnet: 5,698 (87.15%), 768d
image/resnet152: 6,287 (96.16%), 1,000d
image/vgg: 6,287 (96.16%), 4,096d
image/vit_cls: 6,287 (96.16%), 768d
image/vit_avg: 6,287 (96.16%), 768d

Last.fm-2K

Loading: dataset="lfm2k" loads HetRec artist listening counts and metadata from SWAP’s ZIP; optional media vectors are pooled by artist.

Terms: Last.fm / HetRec non-commercial-use conditions; commercial use requires permission. Cite SWAP for its features.

Defaults: Binarize play counts, filter short histories, and retain short artist names. Tag-derived embeddings are interaction-derived, not independent content metadata.

Installed Last.fm-2K preprocessing and split defaults

Dataset / preprocessing

User split

Item split

Leave-last-out

Temporal

Item metadata (union of measured splits)
Last.fm-2K
All feedback
Min text: 0 words
Seed: 42
5/1
1,877 users
17,617 items
Train: 1,277 users
Val: 200 users †
1,070 items / 0.00% cold
Test: 400 users †
1,731 items / 0.00% cold
5/1
1,877 users
17,617 items
Train: 1,877 users
Val: 1,748 users
881 items / 100.00% cold
Test: 1,864 users
1,762 items / 100.00% cold
Not supported
No interaction timestamps
Not supported
No interaction timestamps
Union: 17,617 items
≥1 image: 17,173 (97.48%)
≥10 words: 19 (0.11%)
Precomputed features:
text/minilm: 12,124 (68.82%), 384d
text/mpnet: 12,124 (68.82%), 768d
image/resnet152: 2,314 (13.14%), 1,000d
image/vgg: 2,314 (13.14%), 4,096d
image/vit_cls: 2,314 (13.14%), 768d
image/vit_avg: 2,314 (13.14%), 768d
audio/vggish: 2,364 (13.42%), 128d
audio/whisper: 2,364 (13.42%), 512d
Text: interaction-derived tags
Media: pooled by artist ID

Using item embeddings

SWAP JSON features are opt-in, checksum-verified imports; they do not change the interaction split. Other datasets use user-computed vectors in the same checkpoint format. See Using item embeddings for import, storage, and missing-feature handling; original dataset restrictions still apply.

Precomputed feature names

Modality

Feature names

Datasets

Text

text/minilm, text/mpnet

ML-1M, DBbook, Last.fm-2K

Image

image/resnet152, image/vgg, image/vit_cls, image/vit_avg

ML-1M, DBbook, Last.fm-2K

Audio

audio/vggish, audio/whisper

ML-1M, Last.fm-2K

Video

video/i3d, video/r2p1d

ML-1M