This repository contains the content-based recommendation dataset used in the paper TF-DCon. It contains four types of data:
- original full data
- random sampled data
- majority sampled data
- TF-DCon condensed data
All four types of data share the same user history and test set.
Our experiments are conducted on three widely used datasets:
- MovieLens-100K (movie recommendation)
- Goodreads (book recommendation)
- MIND (news recommendation)
We also provide GPT-generated condensed content and user interests in the raw_data folder.