/KuaiRec

KwaiRec: A Fully-observed Dataset for Recommender Systems.

Primary LanguageJupyter NotebookCreative Commons Attribution Share Alike 4.0 InternationalCC-BY-SA-4.0

KuaiRec: A Fully-observed Dataset for Recommender Systems (Density: Almost 100%)

LICENSE

KuaiRec is a real-world dataset collected from the recommendation logs of the video-sharing mobile app Kuaishou. For now, it is the first dataset that contains a fully observed user-item interaction matrix. For the term "fully observed", we mean there are almost no missing values in the user-item matrix, i.e., each user has viewed each video and then left feedback.

The following figure illustrates the user-item matrices in traditional datasets and KuaiRec.

kuaidata

With all user's preference known, KuaiRec can used in offline evaluation (i.e., offline A/B test) for recommendation models. It can benefit lots of research directions, such as unbiased recommendation, interactive/conversational recommendation, or reinforcement learning (RL) and off-policy evaluation (OPE) for recommendation.

If you use it in your work, please cite our paper: LINK PDF

@inproceedings{gao2022kuairec,
  author = {Chongming Gao and Shijun Li and Wenqiang Lei and Jiawei Chen and Biao Li and Peng Jiang and Xiangnan He and Jiaxin Mao and Tat-Seng Chua},
  title = {KuaiRec: A Fully-observed Dataset and Insights for Evaluating Recommender Systems},
  booktitle = {Proceedings of the 31st ACM International Conference on Information and Knowledge Management},
  series = {CIKM '22},
  location = {Atlanta, GA, USA},
  url = {https://doi.org/10.1145/3511808.3557220},
  doi = {10.1145/3511808.3557220},
  numpages = {11},
  year = {2022}
}

This repository lists the example codes in evaluating conversational recommendation as described in the paper.

We provide some simple statistics of this dataset here . It is generated by Statistics_KuaiRec.ipynb. You can do it online at Google Colab colab.


News!

2022.05.16: We update the dataset to version 2.0. We made the following changes:

  • We removed the unused video ID=1225 from all tables having the field video_id and re-index the rest videos, i.e., ID = ID - 1 if ID > 1225.
  • We added two tables to enhance the side information for users and videos, respectively. See 4.item_daily_features.csv and 5. user_features.csv under the data description section for details.

Download the data

We provides several options to download this dataset:

Option 1. Download via the "wget" command.

 wget https://chongming.myds.me:61364/data/KuaiRec.zip --no-check-certificate
 unzip KuaiRec.zip

Option 2. Download manually throughs the following links:

The script loaddata.py provides a simple way to load the data via Pandas in Python.


Data Descriptions

KuaiRec contains millions of user-item interactions as well as the side information including the item categories and a social network. Six files are included in the download data:

KuaiRec
├── data
│   ├── big_matrix.csv          
│   ├── small_matrix.csv
│   ├── social_network.csv
│   ├── user_features.csv
│   ├── item_daily_features.csv
│   └── item_categories.csv

The statistics of the small matrix and big matrix in KuaiRec.

#Users #Items #Interactions Density
small matrix 1,411 3,327 4,676,570 99.6%
big matrix 7,176 10,728 12,530,806 16.3%

Note that the density of small matrix is 99.6% instead of 100% because some users have explicitly indicated that they would not be willing to receive recommendations from certain authors. I.e., They blocked these videos.

1. Descriptions of the fields in big_matrix.csv and small_matrix.csv.

Field Name: Description Type Example
user_id The ID of the user. int64 0
video_id The ID of the viewed video. int64 3650
play_duration Time of video viewing of this interaction (millisecond). int64 13838
video_duration Time of this video (millisecond). int64 10867
time Human-readable date for this interaction str "2020-07-05 00:08:23.438"
date Date of this interaction int64 20200705
timestamp Unix timestamp float64 1593878903.438
watch_ratio The video watching ratio (=play_duration/video_duration) float64 1.273397

The "watch_ratio" can be deemed as the label of the interaction. Note: there is no "like" signal for this dataset. If you need this binary signal in your scenarios, you can create it yourself. E.g., like = 1 if watch_ratio > 2.0.

2. Descriptions of the fields in social_network.csv

Field Name: Description Type Example
user_id The ID of the user. int64 5352
friend_list The list of ID of the friends of this user. list [4202,7126]

3. Descriptions of the fields in item_categories.csv.

Field Name: Description Type Example
video_id The ID of the video. int64 1
feat The list of tags of this video. list [27,9]

4. Descriptions of the fields in item_daily_features.csv. (Added on 2022.05.16)

Field Name: Description Type Example
video_id The ID of the video. int64 3784
date Date of the statistics of this video. int64 20200730
author_id The ID of the author of this video. int64 441
video_type Type of this video (NORMAL or AD). str "NORMAL"
upload_dt Upload date of this video. str "2020-07-08"
upload_type The upload type of this video. str "ShortImport"
visible_status The visible state of this video on the APP now. str "public"
video_duration The time duration of this duration (in millisecond). float64 17200.0
video_width The width of this video on the server. int64 720
video_height The height of this video on the server. int64 1280
music_id Background music ID of this video. int64 989206467
video_tag_id The ID of tag of this video. int64 2522
video_tag_name The name of tag of this video. string "祝福"
show_cnt The number of shows of this video within this day (the same with all following fields) int64 7716
show_user_num The number of users who received the recommendation of this video. int64 5256
play_cnt The number of plays. int64 7701
play_user_num The number of users who plays this video. int64 5034
play_duration The total time duration of playing this video (in millisecond). int64 138333346
complete_play_cnt The number of complete plays. complete play: finishing playing the whole video, i.e., #(play_duration >= video_duration). int64 3446
complete_play_user_num The number of users who perform the complete play. int64 2033
valid_play_cnt valid play: play_duration >= video_duration if video_duration <= 7s, or play_duration > 7 if video_duration > 7s. int64 5099
valid_play_user_num The number of users who perform the complete play. int64 3195
long_time_play_cnt long time play: play_duration >= video_duration if video_duration <= 18s, or play_duration >=18 if video_duration > 18s. int64 3299
long_time_play_user_num The number of users who perform the long time play. int64 1940
short_time_play_cnt short time play: play_duration < min(3s, video_duration). int64 1538
short_time_play_user_num The number of users who perform the short time play. int64 1190
play_progress The average video playing ratio (=play_duration/video_duration) int64 0.579695
comment_stay_duration Total time of staying in the comments section int64 467865
like_cnt Total likes int64 659
like_user_num The number of users who hit the "like" button. int64 657
click_like_cnt The number of the "like" resulted from double click int64 496
double_click_cnt The number of users who double click the video. int64 163
cancel_like_cnt The number of likes that are cancelled by users. int64 15
cancel_like_user_num The number of users who cancel their like. int64 15
comment_cnt The number of comments within this day. int64 13
comment_user_num The number of users who comment this video. int64 12
direct_comment_cnt The number of direct comments (depth=1). int64 13
reply_comment_cnt The number of reply comments (depth>1). int64 0
delete_comment_cnt The number of deleted comments. int64 0
delete_comment_user_num The number of users who delete their comments. int64 0
comment_like_cnt The number of comment likes. int64 2
comment_like_user_num The number of users who like the comments. int64 2
follow_cnt The number of increased follows from this video. int64 151
follow_user_num The number of users who follow the author of this video due to this video. int64 151
cancel_follow_cnt The number of decreased follows from this video. int64 0
cancel_follow_user_num The number of users who cancel their following of the author of this video due to this video. int64 0
share_cnt The times of sharing this video. int64 1
share_user_num The number of users who share this video. int64 1
download_cnt The times of downloading this video. int64 2
download_user_num The number of users who download this video. int64 2
report_cnt The times of reporting this video. int64 0
report_user_num The number of users who report this video. int64 0
reduce_similar_cnt The times of reducing similar content of this video. int64 2
reduce_similar_user_num The number of users who choose to reduce similar content of this video. int64 2
collect_cnt The times of adding this video to favorite videos. int64 0
collect_user_num The number of users who add this video to their favorite videos. int64 0
cancel_collect_cnt The times of removing this video from favorite videos. int64 0
cancel_collect_user_num The number of users who remove this video from their favorite videos int64 0

5. Descriptions of the fields in user_features.csv (Added on 2022.05.16)

Field Name: Description Type Example
user_id The ID of the user. int64 0
user_active_degree In the set of {'high_active', 'full_active', 'middle_active', 'UNKNOWN'}. str "high_active"
is_lowactive_period Is this user in its low active period int64 0
is_live_streamer Is this user a live streamer? int64 0
is_video_author Has this user uploaded any video? int64 0
follow_user_num The number of users that this user follows. int64 5
follow_user_num_range The range of the number of users that this user follows. In the set of {'0', '(0,10]', '(10,50]', '(100,150]', '(150,250]', '(250,500]', '(50,100]', '500+'} str "(0,10]"
fans_user_num The number of the fans of this user. int64 0
fans_user_num_range The range of the number of fans of this user. In the set of {'0', '[1,10)', '[10,100)', '[100,1k)', '[1k,5k)', '[5k,1w)', '[1w,10w)'} str "0"
friend_user_num The number of friends that this user has. int64 0
friend_user_num_range The range of the number of friends that this user has. In the set of {'0', '[1,5)', '[5,30)', '[30,60)', '[60,120)', '[120,250)', '250+'} str "0"
register_days The days since this user has registered. int64 107
register_days_range The range of the registered days. In the set of {'15-30', '31-60', '61-90', '91-180', '181-365', '366-730', '730+'}. str "61-90"
onehot_feat0 An encrypted feature of the user. Each value indicate the position of "1" in the one-hot vector. Range: {0,1} int64 0
onehot_feat1 An encrypted feature. Range: {0, 1, ..., 7} int64 1
onehot_feat2 An encrypted feature. Range: {0, 1, ..., 29} int64 17
onehot_feat3 An encrypted feature. Range: {0, 1, ..., 1075} int64 638
onehot_feat4 An encrypted feature. Range: {0, 1, ..., 11} int64 2
onehot_feat5 An encrypted feature. Range: {0, 1, ..., 9} int64 0
onehot_feat6 An encrypted feature. Range: {0, 1, 2} int64 1
onehot_feat7 An encrypted feature. Range: {0, 1, ..., 46} int64 6
onehot_feat8 An encrypted feature. Range: {0, 1, ..., 339} int64 184
onehot_feat9 An encrypted feature. Range: {0, 1, ..., 6} int64 6
onehot_feat10 An encrypted feature. Range: {0, 1, ..., 4} int64 3
onehot_feat11 An encrypted feature. Range: {0, 1, ..., 2} int64 0
onehot_feat12 An encrypted feature. Range: {0, 1} int64 0
onehot_feat13 An encrypted feature. Range: {0, 1} int64 0
onehot_feat14 An encrypted feature. Range: {0, 1} int64 0
onehot_feat15 An encrypted feature. Range: {0, 1} int64 0
onehot_feat16 An encrypted feature. Range: {0, 1} int64 0
onehot_feat17 An encrypted feature. Range: {0, 1} int64 0