Resource recommendation vs privacy enhancement
•Download as PPT, PDF•
1 like•400 views
Social tagging has opened new possibilities for applications interoperability on the semantic web, while at the same time posing new privacy treats. Recommendation and information filtering systems in fact predict users preferences, providing personalized content to their users, but also exposing their profiles to possible privacy attacks. Tag suppression and forgery are Privacy Enhancing Techniques that protect users privacy to a certain extent, at the loss of semantic accuracy loss, or in other words privacy gain at the expenses of utility loss. The impact of tag suppression and forgery to content-based recommendation is hence investigated in a real world application scenario.
Recommended
More Related Content
Similar to Resource recommendation vs privacy enhancement
Similar to Resource recommendation vs privacy enhancement (20)
More from Silvia Puglisi
Recently uploaded
Recently uploaded (20)
Resource recommendation vs privacy enhancement
- 1. 1/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Silvia Puglisi silvia.puglisi@upc.edu “Research Seminar” Master in Telematics Engineering-UPC On Content-Based Recommendation and Users Privacy in Social Tagging Systems Silvia Puglisi Barcelona, UPC, 2013
- 2. 2/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Social tagging is the activity that allows users to assign keywords (tags) to web based resources. What is social tagging?
- 3. 3/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Tagging and tags Tag: a label attached to someone or something for identification or other information
- 4. 4/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Scenario Social tagging enables semantic interoperability in web applications. Recommendation and information filtering systems have been developed to predict users preferences. Users hence reveal their personal preferences on social tagging platforms. Privacy enhancing techniques (PET) have been developed to protect user privacy to a certain extent, at the expense of semantic loss.
- 5. 5/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Objective Using as starting point research done in the field of recommendations systems [1] and PET [2]. The objective of this study is evaluate the impact of two PET, tag forgery and suppression, on the performance of a recommendation system, on real world application data. [1] Bellogín, Alejandro, Iván Cantador, and Pablo Castells. "A comparative study of heterogeneous item recommendations in social systems." Information Sciences (2012) [2] Parra-Arnau, Javier, David Rebollo-Monedero, and Jordi Forné. "A privacy-protecting architecture for collaborative filtering via forgery and suppression of ratings." Data Privacy Management and Autonomous Spontaneus Security (2012): 42-57.
- 6. 6/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Dataset Considering different social bookmarking platform, Delicious was identified as a representative system of an application rich in collaborative tagging information. Delicious is a social bookmarking platform for web resources. The dataset containing Delicious data was obtained from the ones publicly available at the 2nd International Workshop on Information Heterogeneity and Fusion in Recommender Systems.
- 7. 7/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Delicious
- 8. 8/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Modelling the User/Item Profile The simplest approach to model users and items is to count the number of times a tag has been used: •By a user to annotate different items in the same category. •Or by the community to annotate the item. The user/item profile is then described as a histogram of the relative frequencies of tags within a predefined set of categories of interest.
- 9. 9/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Histogram of a user profile
- 10. 10/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Privacy Metric The Kullback-Leibler (KL) divergence has been adopted as privacy criteria, following the perspective of Jaynes’ rationale on entropy maximization methods. Since the KL divergence may be regarded as a generalization of entropy of a distribution, relative to another, it is often referred to as relative entropy. D(p || q) = Ep log p(x) q(x) = p(x)log x ∑ p(x) q(x)
- 11. 11/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Utility Metric A measure of how an item is useful for a certain user is needed. We could convey that an item is useful if its profile is somehow similar to the user profile. Hence we need a measure of similarity. Content based recommender models are defined as similarity measures between users and item profiles. This is provided by the cosine-based similarity measure:
- 12. 12/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Performance Metric The recommender system is evaluated considering a content retrieval scenario where a user is provided with a ranked list of N recommended items. The performance metric adopted is hence among the commonly used for ranked list prediction, i.e. precision at top N. In the field of Information Retrieval precision can be defined as the fraction of recommended items that are relevant for a target user.
- 13. 13/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Tag Forgery and Suppression Tag suppression and forgery are privacy enhancing techniques that helps users who tags resources online, from revealing sensible information to a possible attacker.
- 14. 14/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques Tag Forgery and Suppression Rates The tag forgery rate represents the ratio of forged items: The tag suppression rate, is the proportion of items that the user consents to eliminate: ρ ∈ [0,1) σ ∈ [0,1)
- 15. 15/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Techniques The Privacy-Forgery-Suppression Function Consistently the privacy-forgery-suppression function can be defined: P(ρ,σ ) = maxr,s D q +r − s 1+ ρ −σ ÷ ri ≥ 0 ri = ρ i n ∑ qi ≥ si ≥ 0 si = 1 n ∑ σ
- 16. 16/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Evaluation
- 17. 17/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Evaluation Statistics about the dataset Categories 11 Users 1867 Item-Category Tuples 98998 Avg. tags per user 477.75 Items 69226 Avg. Items per Category 81044 Avg. categories per item 1.4 Tags per item 13.06
- 18. 18/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Results Relative Risk Reduction with forgery - Utility 100× Dinit (um || P)− Dρ,σ (um || P) Dinit (um || P)
- 19. 19/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Results Relative Risk Reduction with suppression - Utility
- 20. 20/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Conclusions Tag suppression and forgery are simple privacy enhancing techniques able to protect users privacy at the cost of some semantic loss. This study shows with a simple experimental evaluation, in a real world application scenario, how the performances degradation of a recommender system, is small if compared to the privacy risk reduction offered by the application of these techniques.
- 21. 21/21 Research Seminar. Silvia Puglisi Departament d'Enginyeria Telemàtica Thank you!
Editor's Notes
- In information systems , a tag is a non-hierarchical keyword or term assigned to a piece of information (such as an Internet bookmark , digital image, or computer file ). This kind of metadata helps describe an item and allows it to be found again by browsing or searching. Tags are generally chosen informally and personally by the item's creator or by its viewer, depending on the system. Labeling and tagging are carried out to perform functions such as aiding in classification , marking ownership, noting boundaries, and indicating online identity . They may take the form of words, images, or other identifying marks. An analogous example of tags in the physical world is museum object tagging. In the organization of information and objects, the use of textual keywords as part of identification and classification long predates computers. However, computer based searching made the use of keywords a rapid way of exploring records.
- Tagging has gained wide popularity due to the growth of social networking, photography sharing and bookmarking sites. These sites allow users to create and manage labels (or “tags”) that categorize content using simple keywords. The use of keywords as part of an identification and classification system long predates computers. In the early days of the web keywords meta tags were used by web page designers to tell search engines what the web page was about. Today's tagging takes the meta keywords concept and re-uses it. The users add the tags. The tags are clearly visible, and are themselves links to other items that share that keyword tag. User annotate items that are relevant for them. Tags describe interests, tastes, needs.
- Tagging is popular. Everyone using web or mobile app tags resources online. Many blog systems allow authors to add free-form tags to a post, along with (or instead of) placing the post into categories. For example, a post may display that it has been tagged with baseball and tickets . Each of those tags is usually a web link leading to an index page listing all of the posts associated with that tag. The blog may have a sidebar listing all the tags in use on that blog, with each tag leading to an index page. To reclassify a post, an author edits its list of tags. All connections between posts are automatically tracked and updated by the blog software; there is no need to relocate the page within a complex hierarchy of categories.
- One of the most popular privacy criteria in database anonymisation is k-anonymity. I.e. each combination of key attribute values is shared by at least k records in the set. K-anonymity is vulnerable against similarity attacks. An attacker will be able to compromise user privacy as long as the apparent user profile diverges from a reference probability measure. In probability theory and information theory , the Kullback–Leibler divergence [1] [2] [3] (also information divergence , information gain , relative entropy , or KLIC ) is a non-symmetric measure of the difference between two probability distributions P and Q . Specifically, the Kullback–Leibler divergence of Q from P , denoted D KL ( P || Q ), is a measure of the information lost when Q is used to approximate P : [4] KL measures the expected number of extra bits required to code samples from P when using a code based on Q , rather than using a code based on P . Typically P represents the "true" distribution of data, observations, or a precisely calculated theoretical distribution. The measure Q typically represents a theory, model, description, or approximation of P . Although it is often intuited as a metric or distance, the KL divergence is not a true metric — for example, it is not symmetric: the KL from P to Q is generally not the same as the KL from Q to P . However, its infinitesimal form, specifically its Hessian, is a metric tensor: it is the Fisher information metric.
- Cosine similarity is a measure of similarity between two vectors of an inner product space that measures the cosine of the angle between them. The cosine of 0° is 1, and it is less than 1 for any other angle. It is thus a judgement of orientation and not magnitude: two vectors with the same orientation have a Cosine similarity of 1, two vectors at 90° have a similarity of 0, and two vectors diametrically opposed have a similarity of -1, independent of their magnitude. Cosine similarity is particularly used in positive space, where the outcome is neatly bounded in [0,1].
- Two possible strategies can be contemplated for the user: a mixed strategy, where forgery and suppression are used in conjunction, a pure strategy , where either forgery or suppression are applied. Only the pure strategy is going to be evaluated for the purpose of this study. q is introduced as the probability distribution of the known items of a particular user. This is the probability distribution capturing the actual preferences of the user.