De-anonymizing Users Across Rating Datasets via Record Linkage and Quasi-Identifier Attacks

This implementation follows the steps outlined in the proposed approach:

1. The preprocess_data function filters out users with fewer than a specified number of ratings and normalizes the rating values between 0 and 1.
2. The extract_quasi_identifiers function extracts the rating vectors for each user as quasi-identifiers.
3. The compute_similarities function computes the pairwise cosine similarities between rating vectors of users across the two datasets.
4. The perform_record_linkage function performs record linkage using the Fellegi-Sunter model from the recordlinkage library, taking the similarities and probabilities as input.