Datasets:

Modalities:
Tabular
Formats:
parquet
ArXiv:
Libraries:
Datasets
pandas
License:
Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
uid
uint32
timestamp
uint32
item_id
uint32
is_organic
uint8
10
15,929,455
6,928,395
1
20
18,111,700
6,417,612
0
20
18,111,700
9,362,291
0
20
18,111,945
6,983,647
1
30
106,910
9,270,848
1
30
146,130
6,185,100
1
30
146,130
5,021,833
1
30
673,635
611,534
1
30
852,920
3,805,327
1
30
1,438,455
8,297,002
1
30
1,446,270
2,347,953
1
30
1,454,685
3,715,667
0
30
1,667,925
6,961,827
1
30
1,883,735
4,823,617
1
30
2,029,390
399,811
1
30
2,196,780
6,011,806
1
30
2,393,795
4,610,540
1
30
2,393,835
8,093,245
1
30
2,401,575
721,863
0
30
3,049,440
2,657,039
1
30
3,060,875
1,906,039
1
30
3,329,975
7,175,814
1
30
3,330,000
6,853,370
1
30
3,695,825
3,792,777
1
30
3,695,825
8,549,891
1
30
4,001,290
8,962,495
1
30
4,254,535
593,420
1
30
4,255,160
1,694,706
1
30
4,552,105
2,738,465
1
30
4,603,180
2,612,724
1
30
4,633,740
3,003,160
1
30
4,820,285
7,656,747
1
30
4,820,285
1,529,674
1
30
4,951,580
4,886,017
1
30
4,951,815
684,720
1
30
5,466,570
3,314,114
1
30
5,466,570
771,479
1
30
5,466,570
1,845,537
1
30
5,466,570
6,533,885
1
30
5,507,660
7,336,750
1
30
6,368,890
6,819,516
1
30
6,673,490
4,078,058
1
30
6,767,310
5,198,741
1
30
7,020,170
645,242
1
30
7,020,490
2,737,135
1
30
7,153,885
8,960,892
1
30
7,539,030
5,040,341
0
30
7,667,215
4,534,145
0
30
7,729,080
6,419,137
0
30
7,729,435
8,791,696
0
30
7,732,310
3,890,086
0
30
7,732,675
2,172,147
0
30
7,799,195
846,937
1
30
7,884,640
3,292,875
1
30
7,884,640
9,168,649
1
30
7,885,300
3,159,307
1
30
8,359,975
7,687,183
0
30
8,579,425
4,463,702
1
30
8,933,300
5,578,367
1
30
8,933,340
2,267,298
1
30
8,970,600
7,490,507
1
30
8,970,870
696,835
1
30
9,125,010
4,864,413
1
30
9,125,825
8,232,352
1
30
9,127,500
1,862,391
1
30
9,193,125
5,998,866
0
30
9,373,225
1,332,844
1
30
9,441,015
5,262,221
0
30
9,614,430
8,379,325
1
30
9,631,385
2,473,847
0
30
10,078,345
8,109,341
1
30
10,132,685
7,509,451
1
30
10,144,530
7,267,664
0
30
10,323,335
3,070,695
1
30
10,388,260
2,169,338
0
30
10,734,080
7,767,736
1
30
10,909,925
8,721,143
1
30
10,910,305
3,852,593
1
30
10,950,805
1,325,057
0
30
11,034,490
847,069
1
30
11,434,065
4,441,378
1
30
11,473,350
4,474,559
1
30
11,514,010
3,978,220
1
30
11,735,000
3,978,220
1
30
11,739,040
4,199,147
1
30
11,813,245
3,202,543
1
30
11,945,970
8,420,071
1
30
11,994,510
9,305,160
1
30
12,074,010
290,137
1
30
12,074,010
5,436,843
1
30
12,116,490
9,378,983
1
30
12,161,470
9,201,132
0
30
12,484,735
6,992,647
0
30
12,637,115
2,941,765
0
30
12,637,220
651,447
1
30
12,638,150
5,104,129
1
30
12,723,695
5,284,431
1
30
12,745,990
1,557,849
1
30
12,757,205
2,861,862
1
30
13,154,350
9,164,022
1
End of preview. Expand in Data Studio

Yambda-5B β€” A Large-Scale Multi-modal Dataset for Ranking And Retrieval

Industrial-scale music recommendation dataset with organic/recommendation interactions and audio embeddings

πŸ“Œ Overview β€’ πŸ”‘ Key Features β€’ πŸ“Š Statistics β€’ πŸ“ Format β€’ πŸ† Benchmark β€’ ❓ FAQ

Overview

The Yambda-5B dataset is a large-scale open database comprising 4.79 billion user-item interactions collected from 1 million users and spanning 9.39 million tracks. The dataset includes both implicit feedback, such as listening events, and explicit feedback, in the form of likes and dislikes. Additionally, it provides distinctive markers for organic versus recommendation-driven interactions, along with precomputed audio embeddings to facilitate content-aware recommendation systems.

Key Features

  • 🎡 4.79B user-music interactions (listens, likes, dislikes, unlikes, undislikes)
  • πŸ•’ Timestamps with global temporal ordering
  • πŸ”Š Audio embeddings for 7.72M tracks
  • πŸ’‘ Organic and recommendation-driven interactions
  • πŸ“ˆ Multiple dataset scales (50M, 500M, 5B interactions)
  • πŸ§ͺ Standardized evaluation protocol with baseline benchmarks

About Dataset

Statistics

Dataset Users Items Listens Likes Dislikes
Yambda-50M 10,000 934,057 46,467,212 881,456 107,776
Yambda-500M 100,000 3,004,578 466,512,103 9,033,960 1,128,113
Yambda-5B 1,000,000 9,390,623 4,649,567,411 89,334,605 11,579,143

User History Length Distribution

user history length

user history length log-scale

Item Interaction Count

item interaction count log-scale

Data Format

File Descriptions

File Description Schema
listens.parquet User listening events with playback details uid, item_id, timestamp, is_organic, played_ratio_pct, track_length_seconds
likes.parquet User like actions uid, item_id, timestamp, is_organic
dislikes.parquet User dislike actions uid, item_id, timestamp, is_organic
undislikes.parquet User undislike actions (reverting dislikes) uid, item_id, timestamp, is_organic
unlikes.parquet User unlike actions (reverting likes) uid, item_id, timestamp, is_organic
embeddings.parquet Track audio-embeddings item_id, embed, normalized_embed

Common Event Structure (Homogeneous)

Most event files (listens, likes, dislikes, undislikes, unlikes) share this base structure:

Field Type Description
uid uint32 Unique user identifier
item_id uint32 Unique track identifier
timestamp uint32 Delta times, binned into 5s units.
is_organic uint8 Boolean flag (0/1) indicating if the interaction was algorithmic (0) or organic (1)

Sorting: All files are sorted by (uid, timestamp) in ascending order.

Unified Event Structure (Heterogeneous)

For applications needing all event types in a unified format:

Field Type Description
uid uint32 Unique user identifier
item_id uint32 Unique track identifier
timestamp uint32 Timestamp binned into 5s units.granularity
is_organic uint8 Boolean flag for organic interactions
event_type enum One of: listen, like, dislike, unlike, undislike
played_ratio_pct Optional[uint16] Percentage of track played (1-100), null for non-listen events
track_length_seconds Optional[uint32] Total track duration in seconds, null for non-listen events

Notes:

  • played_ratio_pct and track_length_seconds are non-null only when event_type = "listen"
  • All fields except the two above are guaranteed non-null

Sequential (Aggregated) Format

Each dataset is also available in a user-aggregated sequential format with the following structure:

Field Type Description
uid uint32 Unique user identifier
item_ids List[uint32] Chronological list of interacted track IDs
timestamps List[uint32] Corresponding interaction timestamps
is_organic List[uint8] Corresponding organic flags for each interaction
played_ratio_pct List[Optional[uint16]] (Only in listens and multi_event) Play percentages
track_length_seconds List[Optional[uint32]] (Only in listens and multi_event) Track durations

Notes:

  • All lists maintain chronological order
  • For each user, len(item_ids) == len(timestamps) == len(is_organic)
  • In multi-event format, null values are preserved in respective lists

Benchmark

Code for the baseline models can be found in benchmarks/ directory, see Reproducibility Guide

FAQ

Are test items presented in training data?

Not all, some test items do appear in the training set, others do not.

Are test users presented in training data?

Yes, there are no cold users in the test set.

How are audio embeddings generated?

Using a convolutional neural network inspired by J. Spijkervet et al., 2021.

What's the is_organic flag?

Indicates whether interactions occurred through organic discovery (True) or recommendation-driven pathways (False)

Which events are considered recommendation-driven?

Recommendation events include actions from:

  • Personalized music feed
  • Personalized playlists

What counts as a "listened" track or Listen+Listen_+?

A track is considered "listened" if over 50% of its duration is played.

Downloads last month
6,335