A Fast Recommendation Algorithm for Data Streams Based on Differential Privacy

Recommendation systems are an essential tools for people to access interesting information from vast amounts of data and have a wide range of applications. However, current recommendation algorithms encounter persistent challenges, including significant time overhead and limitations in safeguarding user privacy effectively. These challenges stem from their reliance on static clustering algorithms to compress data streams. To tackle these issues, we introduce the Fast Recommendation Algorithm based on Differential Privacy (FRDP) for data streams. FRDP leverages the growing learning vector quantization technique for clustering and compressing data streams in a single pass. Additionally, FRDP dynamically introduces noise based on clustering error to protect individuals' privacy while using differentially private clustering data for recommendation analysis. Finally, we compare the FRDP algorithm with the best existing recommendation algorithms based on two real datasets. Extensive experimental results demonstrate that, while maintaining similar recommendation accuracy under the same privacy protection constraints, the FRDP algorithm achieves substantial time savings, with an average reduction in running time ranging from 55.82% to 59.89%.