Developing a Secured Recommender System in Social Semantic Network

Developing a Secured Recommender System in Social Semantic Network

• 1.
  Developing a SecureRecommender System in Social Semantic Network Master Thesis Proposal Supervised by: Prof. Dr. Mostafa Aref Presented by: Tamer Rezk Ibrahim 2013 Dr. Khalid El Bahnasy
• 2.
  - 2 - Agenda Motivation  Goal  Recommender System Types – Content-based recommendation – Collaborative filtering based recommendation – Hybrid  Semantic Web in RS  Recommender System Attacks  Related Work  References
• 3.
  - 3 - Motivation What is a Recommender System? – Recommender systems provide a way for information filtering that attempts to present information that are likely of interest to the user.  Why Using Recommender System? – Enhances user experience  Assists users in finding information  Reduces search and navigation time – Increases productivity – Increases credibility – Mutually beneficial proposition  The main problems with RS? – RS has a big problem in user profile injection, a lot of fake accounts has been created to inject target user, so we try to use algorithm to filter these accounts.
• 4.
  - 4 - Goal Our goal is to develop a secure system to prevent profile attacks to enhance relevant recommendations without fake contents.
• 5.
  - 5 - RecommenderSystem Types Figure Sited from : “Tag Based Social Recommender System(RS)” by Aditi Gupta
• 6.
  - 6 - Content-basedRecommendation  Content based RS – Recommend items similar to those users preferred in the past. – Recommendations are based on information on the content of items rather than on other users’ opinions. – Ex. In a movie recommendation application, a movie may be represented by such features as specific actors, director, subject matter, etc.  Pros. – No need for data on other users. – Able to recommend to users with unique tastes. – Able to recommend new and unpopular items  Cons. – Not all content is well represented by keywords, Ex. Images – Unrated items not shown – New user: No history available
• 7.
  - 7 - CollaborativeRS  Collaborative based RS – Use other users’ recommendations (ratings) to judge item’s utility – Key is to find users/user groups whose interests match with the current user – More users, more ratings: better results  Pros. – Extremely powerful and efficient – Very relevant recommendations – Almost all existing commercial recommenders use this approach (e.g. Amazon).  Cons. – New user: No preferences available. – New item: No ratings available.
• 8.
  - 8 - CollaborativeRS  Example: the system needs to make recommendations to customer C  Customer B is very close to C (he has bought all the books C has bought). Book 5 is highly recommended  Customer D is somewhat close. Book 6 is recommended to a lower extent  Customers A and E are not similar at all. Weight=0 Book 1 Book 2 Book 3 Book 4 Book 5 Book 6 Customer A X X Customer B X X X Customer C X X Customer D X X Customer E X X Slide Sited from : “Recommender systems” by Arnaud De Bruyn
• 9.
  - 9 - HybridRS  Hybrid RS – Uses both content based and collaborative filtering. – Introduced to avoid the limitations found in both content and collaborative methods.
• 10.
  - 10 - SemanticWeb in RS  What is Semantic Web? – Semantic web is new evolution of the classical web pages, it add new semantic layer to web pages.  What is the main goal of Semantic Web ? – The main purpose of the Semantic Web is driving the evolution of the current Web by enabling users to find, share, and combine information more easily. – Ontology is an advanced knowledge organization technique as the backbone of the Semantic Web technology. Figure Sited from : “SEMANTIC-ENHANCED WEB-PAGE RECOMMENDER SYSTEMS” by Thi Thanh Sang Nguyen
• 11.
  - 11 - RecommenderSystem Attacks – Recommender systems are vulnerable to profile injection attacks due to their natural openness. In these attacks, some malicious users artificially inject a large number of attack profiles into the systems in order to bias the recommended results to their advantage. Thus, how to effectively and efficiently identify and resist – The profile injection attacks has become an urgent need to resolve problem for the well development and extensive application of Recommender Systems.
• 12.
  - 12 - ExampleProfile Injection  Assume that a memory-based collaborative filtering is used with: – Pearson correlation as similarity measure – Neighborhood size of 1  Only opinion of most similar user will be used to make prediction Item1 Item2 Item3 Item4 … Target Pearson Alice 5 3 4 1 … ? User1 3 1 2 5 … 5 -0.54 User2 4 3 3 3 … 2 0.68 User3 3 3 1 5 … 4 -0.72 User4 1 5 5 2 … 1 -0.02 Sited from : “Attacks on collaborative recommender systems”
• 13.
  - 13 - Exampleprofile injection  Assume that a memory-based collaborative filtering is used with: – Pearson correlation as similarity measure – Neighborhood size of 1  Only opinion of most similar user will be used to make prediction Item1 Item2 Item3 Item4 … Target Pearson Alice 5 3 4 1 … ? User1 3 1 2 5 … 5 -0.54 User2 4 3 3 3 … 2 0.68 User3 3 3 1 5 … 4 -0.72 User4 1 5 5 2 … 1 -0.02 User2 most similar to Alice Sited from : “Attacks on collaborative recommender systems”
• 14.
  - 14 - Exampleprofile injection  Assume that a memory-based collaborative filtering is used with: – Pearson correlation as similarity measure – Neighborhood size of 1  Only opinion of most similar user will be used to make prediction User2 most similar to Alice Attack Item1 Item2 Item3 Item4 … Target Pearson Alice 5 3 4 1 … ? User1 3 1 2 5 … 5 -0.54 User2 4 3 3 3 … 2 0.68 User3 3 3 1 5 … 4 -0.72 User4 1 5 5 2 … 1 -0.02 Attack 5 3 4 3 … 5 0.87 Sited from : “Attacks on collaborative recommender systems”
• 15.
  - 15 - Item1Item2 Item3 Item4 … Target Pearson Alice 5 3 4 1 … ? User1 3 1 2 5 … 5 -0.54 User2 4 3 3 3 … 2 0.68 User3 3 3 1 5 … 4 -0.72 User4 1 5 5 2 … 1 -0.02 Attack 5 3 4 3 … 5 0.87 Example profile injection  Assume that a memory-based collaborative filtering is used with: – Pearson correlation as similarity measure – Neighborhood size of 1  Only opinion of most similar user will be used to make prediction Attack most similar to Alice Attack User2 most similar to Alice Sited from : “Attacks on collaborative recommender systems”
• 16.
  - 16 - Characterizationof Profile Insertion attacks  Attack dimensions – Push attack:  Increase the prediction value of a target item – Nuke attack:  Decrease the prediction value of a target item – Make the recommender system unusable as a whole  No technical difference between push and nuke attacks  Finally: – We will suppose algorithm using clustering to prevent attacks for RS shillers.
• 17.
  - 17 - RelatedWork 1. From Social Network to Semantic Social Network in From Social Network to rom Social Network to Semantic Social Network in Recommender System Recommender System A lot of papers discuss the principle of Recommender System each of them decide which approach will choose, collaborative based recommendation used as approach in Amazon network as example by Khaled Sellami, Mohamed Ahmed-Nacer , Pierre Tiako , Rachid Chelouah & Hubert Kadima[13]. 1. Toward ontology-based personalization of a recommender system in social network Another approach used by LARIS/EISTI to overcome the problems got in Collaborative Filtering, he used Hybrid-filtering approach to overcome new users and new items problems [12].
• 18.
  - 18 - RelatedWork 3. Securing Recommender Systems against Shilling Attacks Using Social-Based Clustering The open nature of collaborative Filtering allows attackers to inject user profile data and force the system to "adapt" in a manner advantageous to them. Previous work has shown both User-based and item-based recommender systems are vulnerable to the segment attack model. [1]
• 19.
  - 19 - References 1.Xiang-Liang Zhang, Tak Man Desmond Lee & Georgios Pitsilis (2013): Securing Recommender Systems against Shilling Attacks Using Social-Based Clustering. 2. Iván Cantador, Alejandro Bellogín, Pablo Castells (2008): Ontology-Based Personalised and Context-Aware Recommendations of News Items. 3. Abeer El-Korany & Salma M. Khatab (2012): Ontology-based Social Recommender System. 4. Pasquale De Meoa, Antonino Nocera a, Giorgio Terracina b, Domenico Ursino (2011): Recommendation of similar users, resources and social networks in a Social Internetworking Scenario. 5. Iván Cantador, Alejandro Bellogín, Pablo Castells (2008): A Multilayer Ontology-based Hybrid Recommendation Model. 6. Xiwang Yang, Yang Guo, and Yong Liu (2011): Bayesian-inference Based Recommendation in Online Social Networks. 7. Xiwang Yanga, Yang Guob, Yong Liua & Harald Steckc. (2013): A Survey of Collaborative Filtering Based Social Recommender Systems. 8. Aleksandra Klašnja-Milićević (2012): Personalized Recommendation Based on Collaborative Tagging Techniques for an E-learning System. 9. Eoin Hurrell (2013): Social Contextuality and Conversational Recommender Systems. 10. Daniel Mican, Loredana Mocean & Nicolae Tomai (2012): Building a Social Recommender System by Harvesting Social Relationships and Trust Scores between Users. 11. LARIS/EISTI Lab., PRES CERGY Univ. & Cergy Pontoise (2010): Toward ontology-based personalization of a recommender system in social network. 12. Khaled Sellami, Mohamed Ahmed-Nacer, Pierre Tiako, Rachid Chelouah & Hubert Kadima (2012): From Social Network to Semantic Social Network in Recommender System.
• 20.
  Questions ?
• 21.
  Thanks