| # Detecting Risky Health Behaviors in TikTok Videos | |
| The "Detecting Risky Health Behaviors in TikTok Videos" dataset is a collection of TikTok video metadata and annotations that aims | |
| to facilitate the development of ML models that can detect and classify risky health behaviors depicted in short-form TikTok videos. | |
| ## Dataset Description | |
| All videos included in this dataset can be found under this Drive Folder: https://drive.google.com/drive/folders/1LUStRDMuE0FDGtHEYJq684_jJOaw7gsQ?usp=drive_link | |
| Our dataset include the following columns: | |
| - ID: A unique identifier for each video entry. | |
| - createTimeISO: The timestamp when the video was created, formatted in ISO 8601 standard. | |
| - commentCount: The total number of comments received on the video. | |
| - diggCount: The number of "diggs" or likes the video has received. | |
| - playCount: The total number of times the video has been played. | |
| - shareCount: The number of times the video has been shared by viewers. | |
| - text: The caption associated with the video. | |
| - webVideoUrl: URL to view the video on TikTok. | |
| - Split: Indicates the dataset division for each entry. | |
| - Risky Behavior: A binary label indicating whether the video depicts risky health behavior (TRUE) or not (FALSE) (Stored in a different csv file). | |
| - Video Name: The name under which the video was saved on Drive (Under the folders Train or Validation or Test). | |
| ## Dataset Collection | |
| The dataset was collected on October 1st, 2024 from TikTok using the Apify TikTok Hashtag Scraper tool. | |
| Here are the detailed steps for data collection: | |
| - Access the Apify TikTok Hashtag Scraper. | |
| - Input the desired hashtags. In this case, the following hashtags were used: #cigarettes, #alcohol, #diet, #workout, #food, #smoking. | |
| - Run the scraper for each hashtag, limiting the number of collected videos to 100 videos per hashtag to ensure a manageable dataset size. | |
| - After scraping, export the data as a CSV file. | |
| I retained the following columns from the scraped data for further analysis: | |
| ### Annotations | |
| #### Annotation process | |
| The annotators of this dataset were asked to watch no more than 30 seconds of each TikTok video, and determine whether the video depicts or mentions a risky health behavior. | |
| - Unrisky Behavior: The video does not show or mention any risky health behavior or the video discourage an unhealthy behavior/ encourage a healthy behavior. This could include either positive or neutral behaviors. | |
| - Risky Behavior: The video shows or mentions smoking (e.g., smoking cigarettes, vaping), or/and the video shows or mentions alcohol consumption (e.g., drinking beer, wine, or any alcoholic beverages), or/and the video shows or mentions unhealthy eating habits (e.g., Overconsumption of Junk/Processed Foods, Excessive Intake of Sugary Beverages, Crash Diets or Extreme Caloric Restriction, Overeating or Encouraging Binge Eating, Neglecting Essential Nutrients, Promoting Harmful Dietary Supplements), or/and the video shows or mentions any other type of risky health behavior (e.g., reckless behavior, drug use, or similar health risks). | |
| ## Codalab Competition Leaderboard | |
| https://codalab.lisn.upsaclay.fr/competitions/20634#learn_the_details |