FMFParking: Federated Matrix Factorization for Parking Lot Recommendation

Chiaoyu Lu,Yushun Fan,Xing Wu,Junqi Zhang

DOI: https://doi.org/10.1109/bigdataservice52369.2021.00021

2021-01-01

Abstract:In recent years, artificial intelligence has developed rapidly and has been applied in various industries. In parking lot services, people have used data mining and analysis such as matrix factorization, decision tree to predict and recommend parking lots for users. However, with the rise of user privacy awareness and the promulgation of regulations, data acquisition and use rights have been restricted, which results in "data islands" and other issues. We construct a parking recommendation system framework of federated learning based on distributed encryption matrix factorization, called "Federated Matrix Factorization for Parking Lot Recommendation" (FMFParking). In the system, we first build a framework based on the user distributed matrix factorization, and then design a homomorphic encryption federated learning with the framework of distributed matrix factorization. We have proved that this system can improve the security of private data while ensuring accuracy. In this system, each user's personal private data and their whereabouts to the parking lot are protected, which can encourage more users to join this system and provide more privacy-dimensional feature information. With the addition of parking lot data, we can increase the diversity and comprehensiveness of the data, which can make us build a more complete and personalized parking lot recommendation system. This article uses real Electronic Toll Collection (ETC) parking lot data and analyzes the experimental results.