A categorized multimodal TikTok dataset

This dataset encompasses 11242 entries of 5137 unique videos listed between the 31st of July and the 4th of August on the TikTok explore page (https://www.tiktok.com/explore). The page was accessed via a German IP address without being logged in. The data has been collected via the 4CAT Toolkit and the Zeeschuimer browser extension. The dataset contains the category and multimodal embeddings for each video.
Intended Purpose: The dataset is primarily intended for proof-of-concept studies, as a toy dataset to teach or to be used for seminar papers by students. Given the lack of a clear definition for each category by TikTok, the focus of such work might be to explore those definitions or to conduct work with a focus on methods. The multimodal embeddings allow for directly applying unsupervised and supervised machine learning techniques.
Contents: The dataset consists of four zipped .csv files: metadata.zip, text_embeddings.zip, audio_embeddings.zip, video_embedding.zip. For further details, please consult the Data Report (datenbericht_v2.pdf).