PRESERVING PRIVACY IN SEMANTIC-RICH TRAJECTORIES OF HUMAN MOBILITY Anna Monreale, Roberto Trasarti, Dino Pedreschi, Chiara...
How the story begins… Semantic trajectories represent the important places visited by people This information can be priva...
Semantic Trajectories <ul><li>Availability of trajectory data increases </li></ul><ul><li>From raw trajectories to new for...
Semantic Trajectory and Privacy <ul><li>Data owner should not reveal personal sensitive information </li></ul><ul><li>Dis...
Semantic Trajectories Analysis and Privacy Issues <ul><li>Analyzing datasets of semantic trajectories may cause privacy is...
Semantic Trajectories Analysis and Privacy Issues <ul><li>k-anonymity is not enough for a robust protection </li></ul><ul>...
The Privacy Framework <ul><li>Anonymizes dataset of semantic trajectories </li></ul><ul><li>Based on semantic generaliz...
Quasi-identifier and Sensitive stops <ul><li>The taxonomy of places </li></ul><ul><ul><li>Represents important places and...
Privacy place taxonomy
Privacy Model <ul><li>Adversary Knowledge : </li></ul><ul><ul><li>how we anonymize the data </li></ul></ul><ul><ul><li>th...
C-Safe Dataset <ul><li>We want to control the probability Prob(SQ, S) </li></ul><ul><li>A dataset ST is said c-safe wrt...
How we can obtain a c-safe dataset? <ul><li>The CAST (C-safe Anonymization of Semantic Trajectories) algorithm guarantees ...
Example (1): The process Consider the following set of sequences, and m=3 and c=0.45: S = { <S1, R2, H1 , R1, C1 , S2>...
Example (2) CostQ CostQ is the number of hops on the tree needed to generalize the sequences of Quasi-identifiers to a co...
Example (2) CostS CostS is the number of hops on the tree needed to generalize the sequence of Sensible in order to obtain...
Example (4): Why is C-safe <ul><li><Station,Place,Entertainment,S2 (H1,C1) > </li></ul><ul><li><Station,Place,Entertainme...
Experiments <ul><li>We found 6225 semantic trajectories with an average length equal to 5.2 stops. </li></ul><ul><li>We ru...
Experiments: Quality of the analysis <ul><li>the coverage coefficient measures how many patterns extracted from the orig...
Experiments: Coverage Coefficient
Experiments: Quality of the analysis <ul><li>Distance coefficient represents the distance in terms of steps in the taxono...
Experiments: Distance Coefficient
Conclusions and Future work <ul><li>Improve the algorithm with better heuristics and that does not consider only groups of...
of 22

Preserving Privacy in Semantic-Rich Trajectories of Human Mobility

The increasing abundance of data about the trajectories ofpersonal movement is opening up new opportunities for analyzing and mining human mobility, but new risks emergesince it opens new ways of intruding into personal privacy.Representing the personal movements as sequences of placesvisited by a person during her/his movements - semantictrajectory - poses even greater privacy threats w.r.t. rawgeometric location data. In this paper we propose a privacy model defining the attack model of semantic trajectorylinking, together with a privacy notion, called c-safety. Thismethod provides an upper bound to the probability of inferring that a given person, observed in a sequence of non-sensitive places, has also stopped in any sensitive location.Coherently with the privacy model, we propose an algorithmfor transforming any dataset of semantic trajectories into ac-safe one. We report a study on a real-life GPS trajectory dataset to show how our algorithm preserves interestingquality/utility measures of the original trajectories, such assequential pattern mining results.
Published on: Mar 4, 2016
Source: www.slideshare.net


Transcripts - Preserving Privacy in Semantic-Rich Trajectories of Human Mobility

  • 1. PRESERVING PRIVACY IN SEMANTIC-RICH TRAJECTORIES OF HUMAN MOBILITY Anna Monreale, Roberto Trasarti, Dino Pedreschi, Chiara Renso KDDLab, Pisa Vania Bogorny Univ. Santa Catarina, Brasile Knowledge Discovery and Delivery Lab (ISTI-CNR & Univ. Pisa) www-kdd.isti.cnr.it ANONIMO MEETING, Pisa, 20,21 settembre 2010 SPRINGL 2010, San Jose, November 2, 2010
  • 2. How the story begins… Semantic trajectories represent the important places visited by people This information can be privacy sensitive! We should find a good generalization of the visited places… preserving semantics! But how? Can we use a taxonomy of places to generalize and find anonymous datasets? Let’s ask help to Anna, Dino and Roberto!
  • 3. Semantic Trajectories <ul><li>Availability of trajectory data increases </li></ul><ul><li>From raw trajectories to new forms of trajectory data with richer semantic information: semantic trajectories </li></ul><ul><li>Semantic trajectories represents moving objects traces as sequences of stops and moves </li></ul><ul><li>A semantic trajectory can be represented as the sequence of stops, e.g. </li></ul><ul><li><Home, Work, ShoppingCenter, Gym> </li></ul>
  • 4. Semantic Trajectory and Privacy <ul><li>Data owner should not reveal personal sensitive information </li></ul><ul><li>Disclosure of personal sensitive information puts the citizen’s privacy at risk. </li></ul><ul><li>Hiding personal identifiers may not be sufficient </li></ul><ul><li>Need for new privacy-preserving DT techniques </li></ul><ul><ul><li>Privacy by Design </li></ul></ul><ul><li>Natural trade-off between privacy quantification and data utility </li></ul><ul><ul><li>Analysis results should not be altered significantly </li></ul></ul><ul><ul><li>Privacy has to be maximized </li></ul></ul>
  • 5. Semantic Trajectories Analysis and Privacy Issues <ul><li>Analyzing datasets of semantic trajectories may cause privacy issues </li></ul><ul><li>A place allows to infer personal sensitive information of an individual </li></ul><ul><li>Example: From the fact that a person has stopped in an oncology clinic , an attacker can derive private personal information about the health of such person. </li></ul>
  • 6. Semantic Trajectories Analysis and Privacy Issues <ul><li>k-anonymity is not enough for a robust protection </li></ul><ul><li>When individuals with similar trajectories stop in the same sensitive place, we can easily infer the individual sensitive information. </li></ul><ul><li>Example: </li></ul><ul><li>#U1 <Park, Restaurant, Oncology Clinic> </li></ul><ul><li>#U2 <Park, Restaurant, Oncology Clinic> </li></ul><ul><li>This dataset is 2-anonymous but the attacker can infer that the user has been to the Oncology Clinic!!! </li></ul>
  • 7. The Privacy Framework <ul><li>Anonymizes dataset of semantic trajectories </li></ul><ul><li>Based on semantic generalization and the notion of c-safety - similar to the notion of l-diversity in relational, tabular data </li></ul><ul><li>It is based on: a taxonomy of places, the notion of quasi identifier places and sensitive places . </li></ul><ul><li>Preserves patterns mining results </li></ul>
  • 8. Quasi-identifier and Sensitive stops <ul><li>The taxonomy of places </li></ul><ul><ul><li>Represents important places and their semantic categories in a given domain </li></ul></ul><ul><li>quasi-identifier places: can be used to infer the identity of the user </li></ul><ul><li>sensitive places: can disclose sensitive information about the user </li></ul><ul><li>In general we don’t have an apriori classification since it depends on the application and the context </li></ul>
  • 9. Privacy place taxonomy
  • 10. Privacy Model <ul><li>Adversary Knowledge : </li></ul><ul><ul><li>how we anonymize the data </li></ul></ul><ul><ul><li>the privacy place taxonomy describing the levels of abstraction </li></ul></ul><ul><ul><li>the user U is in the dataset </li></ul></ul><ul><ul><li>a quasi-identifier place sequence SQ visited by the user U </li></ul></ul><ul><li>  Attack Model: </li></ul><ul><ul><li>Given SQ, the attacker builts a set of candidate semantic trajectories containing SQ and tries to infer the sensitive places visited by U . </li></ul></ul><ul><ul><li>We denote by Prob(SQ,S) the probability that, given a quasi-identifier place sequence SQ related to a user U , the attacker infers the sequence of sensitive places S visited by the user. </li></ul></ul>
  • 11. C-Safe Dataset <ul><li>We want to control the probability Prob(SQ, S) </li></ul><ul><li>A dataset ST is said c-safe wrt the place set Q if for every quasi-identifier place sequence SQ, we have that for each set of sensitive place S Prob(SQ,S) ≤ c with c ∈ [0,1]. </li></ul><ul><li>Given a sequence of sensitive places S = s 1 , . . . , s h and a quasi-identifier sequence SQ the probability to infer S is the conditional probability: </li></ul><ul><li>P(SQ,S) = P(S|SQ) </li></ul>
  • 12. How we can obtain a c-safe dataset? <ul><li>The CAST (C-safe Anonymization of Semantic Trajectories) algorithm guarantees that P(S|SQ) ≤ c for each sequence of S and SQ </li></ul><ul><li>While (|S|>0) </li></ul><ul><li>S L = { s  S| length(s) = MaxLength(S)} </li></ul><ul><ul><li>While (|S L | >= m) </li></ul></ul><ul><ul><ul><li>Compute the Cost of all possible group G i of m sequences in S L as: Cost Gi = CostQ Gi + CostS Gi . </li></ul></ul></ul><ul><ul><ul><li>Apply the generalization with the lower Cost storing the results in R. </li></ul></ul></ul><ul><ul><ul><li>Remove Gi from S and S L . </li></ul></ul></ul><ul><ul><li>If (| S L |>0) Cut(S L ); </li></ul></ul>
  • 13. Example (1): The process Consider the following set of sequences, and m=3 and c=0.45: S = { <S1, R2, H1 , R1, C1 , S2> <S3, D1, R1, C1 , S2> <S1, P3, C2 , D2, S2> … }
  • 14. Example (2) CostQ CostQ is the number of hops on the tree needed to generalize the sequences of Quasi-identifiers to a common one. Consider the group: <S1, R2, H1 , R1, C1 , S2> <S3, D1, R1, C1 , S2> <S1, P3, C2 , D2, S2> CostQ = 6 + 6 + 6 = 18 <Station,Place,Entertainment,S2 (H1,C1) > <Station,Place,Entertainment,S2 (C1) > <Station,Place,Entertainment,S2 (C2) >
  • 15. Example (2) CostS CostS is the number of hops on the tree needed to generalize the sequence of Sensible in order to obtain the c-safety. From the generalized group: <Station,Place,Entertainment,S2 (H1,C1) > <Station,Place,Entertainment,S2 (C1) > <Station,Place,Entertainment,S2 (C2) > CostS = 3 The Total Cost of this group is 21 hops, which is the lower combination <Station, Place, H1 , Entertainment, Clinic , S2 > <Station, Place, Entertainment, Clinic , S2> <Station, Place, Clinic , Entertainment, S2>
  • 16. Example (4): Why is C-safe <ul><li><Station,Place,Entertainment,S2 (H1,C1) > </li></ul><ul><li><Station,Place,Entertainment,S2 (C1) > </li></ul><ul><li><Station,Place,Entertainment,S2 (C2) > </li></ul><ul><li>SQ = ⟨Station, Place, Entertainment, S2⟩. </li></ul><ul><li>Probability of crack: P (SQ , H 1 ) = 1/3 <c , P(SQ,C 1 ) = 2/3 > c and P(SQ,C 2 ) = 1/3 <c </li></ul><ul><li>We need to generalize C1 to the higher representation level in the taxonomy: Clinic. </li></ul><ul><li>The probability of C1 become 2/5 < c !!!! </li></ul><ul><li>C-safe dataset: </li></ul><ul><li><Station, Place, H1, Entertainment, Clinic, S2 > </li></ul><ul><li><Station, Place, Entertainment, Clinic, S2> </li></ul><ul><li><Station, Place, Clinic, Entertainment, S2> </li></ul>
  • 17. Experiments <ul><li>We found 6225 semantic trajectories with an average length equal to 5.2 stops. </li></ul><ul><li>We run the sequential pattern algorithm and we measured the quality of the results with two measures: </li></ul><ul><li>the coverage coefficient </li></ul><ul><li>the distance coefficient. </li></ul>The dataset contains trajectories of 17000 moving cars in Milan, in one week, collected through GPS devices.
  • 18. Experiments: Quality of the analysis <ul><li>the coverage coefficient measures how many patterns extracted from the original dataset are covered (have a superclass in the taxonomy) by the patterns extracted in the anonymized dataset </li></ul>
  • 19. Experiments: Coverage Coefficient
  • 20. Experiments: Quality of the analysis <ul><li>Distance coefficient represents the distance in terms of steps in the taxonomy to transform the patterns from the set extracted on the original dataset and the one from the anonymized dataset. </li></ul>
  • 21. Experiments: Distance Coefficient
  • 22. Conclusions and Future work <ul><li>Improve the algorithm with better heuristics and that does not consider only groups of a fixed size. </li></ul><ul><li>More experiments with other mining algorithms </li></ul><ul><li>More utility measures for the evaluation of results </li></ul><ul><li>Another future research direction goes towards the exploitation of c-safe semantic trajectories dataset for semantic tagging of trajectories. How does the anonymization step affect the overall results of a trajectory semantic tagging inference? </li></ul>

Related Documents