Currently, we include MovieLens, Book-Crossing, CiteULike, and XING. We will continue to update appropriate and new datasets.
The statistics of the included datasets are as follows:
| Dataset | Users | Items | Interactions | User Contents | Item Contents |
|---|---|---|---|---|---|
| MovieLens | 6,040 | 3,706 | 1,000,209 | 3,469 | 206 |
| Book-Crossing | 92,107 | 270,170 | 1,031,175 | 2,312 | 416 |
| CiteULike | 5,551 | 16,980 | 204,986 | N/A | 300 |
| 106,881 | 20,519 | 3,856,580 | N/A | 2,738 |
The link (Google Drive) to access the datasets is available at: https://drive.google.com/drive/folders/13JJ25vf5dpFzxe1ITQYONrEIUAsB2ZU8?usp=sharing.
We have provided the code to directly dataset processing. You can process the dataset into training format in the following two steps:
python split.py --dataset [DATASET NAME] --cold_object [user/item]
python convert.py --dataset [DATASET NAME] --cold_object [user/item]In the above scripts, the [DATASET NAME] for --dataset should be replaced by your target dataset name, such as movielens. Then, the [user/item] for --cold_object should be selected as user or item for the user cold-start setting or item cold-start setting.