Tokyo TV Dataset is a benchmark dataset for recommendation system and was introduced in our study [1]. We had collected one month of watch logs of TV programs by residents in Tokyo 23 wards. This dataset also contains surrounding knowledge of each program such as program genres, performers, and broadcasters. So you can utilize those knowledge for recommendation.
Details
Tokyo TV Dataset contains
7,183 watch logs of TV programs by 70 users
- To collect the logs, we employed 70 residents of Tokyo 23 wards who indicated "I watch TV at least two hours every day" in the preliminary survey
- Aked them to keep logs of which TV programs they watched and for how long
- 1: only a little
- 2: more than half of the time
- 3: almost all of the time
- Collection period is 4 weeks from Dec. 1 to 28, 2020
- Only programs broadcasts by 7 major broadcasters in the Tokyo area
579 items (TV programs)
- Periodic programs were consolidated as one single item for simplicity's sake
Three types of attributes for each program
We obtained these attributes mainly from M Data Co., Ltd. For performers, we also collected from Wikidata for more data.
- 11 Genres
- 7 broadcasters
- 9,374 performers (actors/actresses, comedians and other celebrities)
To avoid copyright issues, all the items and attributes are represented by unique IDs not text of their actual name.
File Format
The dataset is provided as a dictionary in a json format. The keys in the dictionary is the user_id (string.)
For each key, it includes a list with following values:
- an number in string format as the user_id.
- an number in string format as the item_id.
- a float number as the score assigned by the user.
- an int value as the timestamp for watching logs.
- a list of int as the creator (TV station) of the program.
- a list of int as the genre of the program.
- a list of int as the characters (people who have performed in the program) in the program.
License
This dataset is licensed under Attribution-NonCommercial-ShareAlike 4.0 International
If you use this dataset for research purpose, please cite our paper[1].
Download
Contact
Yuxun Lu: 
Kosuke Nakamura: 
Ryutaro Ichise: 
Reference
- Yuxun Lu, Kosuke Nakamura, and Ryutaro Ichise. "HyperRS: Hypernetwork-Based Recommender System for the User Cold-Start Problem." IEEE Access, vol. 11, pp. 5453-5463, 2023.