SLIDE 1
Tagvisor: A Privacy Advisor for Sharing Hashtags
Yang Zhang
Joint work with Mathias Humbert, Tahleen Rahman, Cheng-Te Li, Jun Pang and Michael Backes
#hashtag
2
Tagvisor: A Privacy Advisor for Sharing Hashtags Yang Zhang Joint - - PowerPoint PPT Presentation
Tagvisor: A Privacy Advisor for Sharing Hashtags Yang Zhang Joint - - PowerPoint PPT Presentation
Tagvisor: A Privacy Advisor for Sharing Hashtags Yang Zhang Joint work with Mathias Humbert, Tahleen Rahman, Cheng-Te Li, Jun Pang and Michael Backes #hashtag 2 #hashtag 3 #hashtag 4 #hashtag 5 #hashtag #like4like #foodporn
SLIDE 2
SLIDE 3
#hashtag
3
SLIDE 4
#hashtag
4
SLIDE 5
#hashtag
5
SLIDE 6
#hashtag
6
#like4like #foodporn #tbt
SLIDE 7
#hashtag
7
#privacy #locationprivacy
SLIDE 8
#contributions
- Attack: location inference with hashtags
- Defense: Tagvisor, a privacy advisor to mitigate the
privacy threat by hashtags
8
SLIDE 9
#dataset
- Collected through Instagram’s APIs
- New York, Los Angeles, and London
- Hashtags + locations (check-ins)
9
SLIDE 10
#attack
10
[1, 1, 1, 0] [0, 1, 1, 0] [1, 0, 0, 1]
- Bag-of-words for feature representation
- Random forest classifier
- Multiple-class classification, e.g., 498 classes (locations) in New York
- All posts are trained together
#a#b#c #b#c #a#d
SLIDE 11
#attack
11
SLIDE 12
#attack
12
SLIDE 13
#tagvisor
- A privacy advisor for sharing hashtags
- Fool the attacker’s location inferencer (ML classifier)
- Three defense mechanisms
- Hiding
- Replacement
- Generalization (location category)
- Utility: preserving the semantical meaning of hashtags
13
SLIDE 14
#hiding
14
hide #a hide #b hide #c successful attack delete one hashtag (can be more) #a#b#c #b#c #a#c #a#b
SLIDE 15
#utility
15
- Semantical meaning
- Skip-gram, aka word2vec
- Skip-gram over all posts’ hashtags
#a: [3.1, 1.3] #b: [2.5, 1.9] #c: [4.0, 5.1] #a #b #c #a#b#c #a#c #a#b Hashtag vectors d1 d2 d1 d2 #a#b#c #a#c #a#b
SLIDE 16
#replacement
16
- Replace each hashtag with all the possible hashtag
- Search space is too big
- Bound to the most closest hashtags (with word2vec)
- Reduce the search space
- Semantical meaning can be preserved
successful attack #a#b#c
SLIDE 17
#generalization
- Location category from foursquare
- #centralpark -> #park
- Do not apply to all hashtags
- e.g., #tbt #love
17
SLIDE 18
#tagvisor
- Check whether the post’s location is inferred correctly
- If no, then publish
- Else, consider the three defense mechanisms
- Pick the hashtag set with the highest utility
18
SLIDE 19
#tagvisor
19
Obfuscating 2 hashtags is enough!
Obfuscating bounded number of hashtags
SLIDE 20
#conclusion
- First location inference attack with hashtags
- Sharing hashtags is not safe!!!
- A privacy advisor to mitigate this risk
- Minimal risk and maximal utility
- Fit for the real-world setting
20