Format ==================== 1. Image features (Table 1) Format: anonymized_pin_id | 14 Image quality features | 1000 Object features | 4096 Deep learning features (The position of each image quality and object features will be also be provided.) 2. User features (Table 1) Format: anonymized_user_id | board_count | pin_count | like_count | follower_count | following_count | 32 user category preference (The object preference can be calculated from object features in 2.) 3. Repin activity (ranked by creation time, ground truth and training data for machine learning): Format: anonymized_user_id | anonymized_pin_id | category | anonymized_board_id Additional data ================ We have anonymised the pin_ids and the user_ids to preserve privacy. To allow researchers to construct other features for the same image, we can provide a link to the image in the Pinterest media cache, which is not traceable to the original image pinned or the user who pinned it. This would be in the format: anonymized_pin_id | Image URL in Pinterest media cache If you need this, please explicitly ask for this in your request. We reiterate that a key condition of sharing this dataset is that we explicitly forbid any attempt to link these cached images to Pinterest pins or user_ids. This condition applies to the image URLs too.