mie-lab / topology_privacy

How privacy-preserving are graph representations of mobiltiy

Influence of tracking duration on the privacy of individual mobility graphs

This is the open-source code for our paper presented at LBS 2022, the 17th International Conference on Location Based Services in Munich. Our paper analyzes how privacy-preserving are graph representations of mobility.

This repository includes all source code used to produce the results from the paper; however, the dataset can not be published due to data privacy.

Installation

The following packages are required to run this code:

• trackintel
• graph_trackintel
• pandas, numpy, scipy
• matplotlib, seaborn
• networkx
• psycopg2, sqlalchemy

Analysis

Our analysis comprises the following steps:

1. precompute features A script that precomputes the graph features used in the publication.
   • transition_feats: The distribution of transition weights over the 20 most popular trips.
   • shortest_path_feats: The distribution of shortest-path lengths in the graph.
   • centrality_feats: The betweenness centrality of a node denotes its centrality with respect to other nodes.
   • in_degree_feats: Distribution of (unweighted) node in-degrees.
   • out_degree_feats: Similar to the in-degree, the distribution of out-degrees over the 20 locations with the highest out-degree is computed.
2. create cross join table
   • Loads feature table
   • Combines all pairs of subsequent time periods for reidentification tests
3. compute similarity
   • Loads the cross-joined pairs
   • Computes similarity with several metrics for each pair
   • Writes the result to the database
4. rank users
   • Loads the similarities
   • For each duration-bin combination, rank the users from the pool by their distance to the current user
   • Write the rank of the matched user to the database
5. fill_matrix:
   • Loads similarities for all combinations (all users & time period bins)
   • Computes reidentification accuracy for all time-bin combinations (over users)
   • Computes reciprocal ranks
6. visualization: Functions for visualizing / summarizing all results reported in the paper, namely
   • The reidentification accuracies (Figure 2)
     • How much is the reidentification top-k accuracy for differnt tracking periods
   • Regression analysis (Table 1)
     • performs the regression analysis to evaluate the effect of pool- and test-user tracking duration on the matching performance
   • Feature analysis (Table 2)
     • which features improve user reidentification performance?
   • Privacy loss analysis (Figure 3)
     • What is the privacy loss due to reidentifying users?
   • Intra and inter user differences (Figure 4)
     • Is the variance explained by differences between users or by differences between time periods?