Social Web applications have emerged as powerful applications for Internet users allowing them to freely contribute to the Web content, organize and share information, and utilize the collective knowledge of others for discovering new topics, resources and new
friends. While social Web applications such as social tagging systems have many benefits, they also present several challenges due to their open and adaptive nature. The amount of user generated data can be extremely large and since there is not any controlled vocabulary or hierarchy, it can be very difficult for users to find the information that is of their interest.
In addition, attackers may attempt to distort the system’s adaptive behavior by inserting erroneous or misleading annotations, thus altering the way in which information is presented to legitimate users. This thesis utilizes data mining and machine learning techniques to address these problems. In particular, we design and develop recommender systems to aid the user in contributing to the Social Semantic Web. In addition, we study intelligent techniques to
combat attacks against social tagging systems. In our work, we first propose a framework that maps domain properties to recommendation technologies. This framework provides a systematic approach to find the appropriate recommendation technology for addressing a
given problem in a specific domain. Second, we improve existing graph-based approaches for personalized tag recommendation in folksonomies. Third, we develop machine learning algorithms for recommendation of semantic relations to support continuous ontology development in a social semanticWeb environment. Finally, we introduce a framework to analyze different types of potential attacks against social tagging systems and evaluate their impact
on those systems.
Contents
1 Introduction
1.1 Motivations and Research Questions
1.2 Thesis overview and contributions
1.2.1 Matching Recommendation Technology and Domains
1.2.2 Improving Link Analysis for Tag Recommendation in Social Tagging Systems
1.2.3 Using Recommender Systems for Continuous Ontology development
1.2.4 Combating Attacks Against Social Tagging Environments
2 Background
2.1 Recommender Systems
2.2 Recommendation Algorithms
2.2.1 Collaborative Filtering
2.2.2 Content-based Recommendation
2.2.3 Knowledge-based Recommendation
2.2.4 Other Types of Recommendation
2.3 Social Tagging Systems
2.3.1 Challenges in Social Tagging Systems
2.3.2 Formalization of Folksonomy
2.4 Related Work in Social Tagging Research
2.4.1 Characterize Social Tagging Systems and User Motivations
2.4.2 Resource Recommendation and Personalized Search
2.4.3 Tag Recommendation
2.4.4 User Recommendation
2.4.5 Emergent Ontologies from Folksonomies
2.5 Chapter Summary
3 Matching Recommendation Technologies and Domains
3.1 Introduction
3.2 Related Work
3.3 Knowledge Sources
3.3.1 Individual Knowledge
3.3.2 Social Knowledge
3.3.3 Content
3.4 Recommendation types
3.5 Domain Properties
3.5.1 Heterogeneity
3.5.2 Risk
3.5.3 Churn
3.5.4 Interaction Style
3.5.5 Preference stability
3.5.6 Inscrutability
3.6 Mapping Knowledge Sources to Domain Properties
3.6.1 Individual
3.6.2 Social Knowledge
3.6.3 Content and Domain Knowledge
3.7 Mapping Domains to Technologies
3.7.1 Algorithms
3.8 Sample Recommendation Domains
3.9 Domain of This Thesis: Social Web
3.10 Conclusion
3.11 Acknowledgements
4 Improving Link Analysis for Tag Recommendation in Folksonomies
4.1 Introduction
4.2 Related Work
4.3 Background on PageRank Algorithm
4.3.1 Folksonomy-Adapted PageRank
4.3.2 FolkRank
4.3.3 Graph-Based Tag Recommendation in Folksonomies
4.4 A weighted Directed Graph Model for Folksonomies
4.5 Experimental Evaluation
4.5.1 Datasets
4.5.2 Experimental Results
4.5.3 Discussion
4.6 Conclusion and Future Work
4.7 Acknowledgment
5 Using Recommender Systems to Support Continuous Ontology Development
5.1 Introduction
5.2 Related Work
5.3 Application Scenarios
5.3.1 Floyd
5.3.2 SOBOLEO
5.3.3 Wikipedia
5.4 Algorithms
5.4.1 Algorithm 1: Recommendation of Super-Concepts Without An Existing Seed Hierarchy
5.4.2 Algorithm 2: Recommendation of Super-Concepts by Learning From an Existing Concept Hierarchy
5.4.3 Algorithm 3: Hybrid Recommendation
5.5 Evaluation
5.5.1 Data Sets
5.5.2 Evaluation Methodology
5.5.3 Experimental Metrics
5.6 Experimental Results
5.6.1 Experimental Results from Algorithm 1
5.6.2 Experimental Results from Algorithm 2
5.6.3 Experimental Results from Algorithm 3
5.6.4 Expert Evaluation
5.6.5 Discussion
5.7 Conclusion and Future Work
5.8 Acknowledgment
6 Combating Attacks Against Social Tagging Environments
6.1 Introduction
6.2 Related Work
6.3 Navigation Channels in Social Tagging Systems
6.4 Attack Dimensions
6.4.1 Attack Types
6.5 Retrieval Algorithms
6.6 Evaluating Impact of Attacks
6.6.1 Measuring the Local Impact of Attack
6.6.2 Measuring the Global Impact of Attack
6.7 Experimental Results
6.7.1 Experimental Setup
6.7.2 Overload Attack
6.7.3 Piggyback Attack
6.7.4 Co-occurrence Attack (Tag Push)
6.7.5 Comparison Of Attack Impact On Different Data Sets
6.7.6 Comparison of Different Attack Types
6.7.7 Comparison of Local Impact
6.8 Discussion
6.9 Conclusion and Future Work
6.10 Acknowledgment
7 Conclusion
7.1 Answer to Research Questions
7.1.1 Objective 1: Matching Recommendation Technology and Domains
7.1.2 Objective 2: Improving Link Analysis for Tag Recommendation in Social Tagging Systems
7.1.3 Objective 3: Using Recommender Systems for Continuous Ontology Development
7.1.4 Objective 4: Combating Attacks Against Social Tagging Systems
7.2 Summary of Contributions
7.2.1 Conceptual Contributions
7.2.2 Algorithm Development
7.2.3 Empirical Contributions
7.3 Future Directions
7.3.1 Explore Other Recommendation Tasks
7.3.2 Network Evolution and Attacks
7.3.3 Scalability And Real-time Analysis
Research Objectives and Themes
This thesis investigates the application of data mining and machine learning algorithms to social Web applications, specifically targeting social tagging systems. The primary research goal is to improve the usability and security of these systems by developing recommender systems that assist users in making meaningful contributions and by creating frameworks to analyze and defend against potential adversarial attacks that aim to manipulate system behavior.
- Matching recommendation technologies to specific application domains.
- Improving link analysis techniques for personalized tag recommendation in folksonomies.
- Developing machine learning algorithms to support the continuous, collaborative evolution of ontologies.
- Modeling and analyzing the impact of profile injection attacks on social tagging environments.
- Defining evaluation frameworks for the local and global impact of such attacks.
Excerpt from the Book
2.3.1 Challenges in Social Tagging Systems
Despite the many benefits offered by folksonomies, they also present unique challenges. In this section, we briefly discuss some of the major challenges including ambiguity, redundancy, and attacks against social tagging systems.
Ambiguity and Redundancy
Most collaborative tagging applications allow the user to describe a resource with any tag they choose. As a result they contain numerous ambiguous and redundant tags.
Ambiguous tags have multiple meanings. A tag may have different word senses; “apple” can refer to the company or to the fruit. Names may also result in ambiguity; “paris” might mean the city or the celebrity. Subjective tags such as “cool” can result in ambiguity since different users have contradictory notions of what constitutes cool. Finally, overly vague tags such as “tool” can mean gardening implements or software packages. Ambiguous tags can impede users as they navigate the system or burden the user with unwanted recommendations.
Summary of Chapters
1 Introduction: Provides an overview of social Web applications, outlines the research objectives, and defines the central research questions regarding system usability and security.
2 Background: Reviews fundamental concepts in recommender systems, social tagging, and identifies the core challenges like ambiguity and system attacks.
3 Matching Recommendation Technologies and Domains: Introduces a taxonomy for mapping different recommendation algorithms to specific domain characteristics and knowledge sources.
4 Improving Link Analysis for Tag Recommendation in Folksonomies: Proposes a weighted directed graph model for folksonomies to improve tag recommendations using an adapted PageRank algorithm.
5 Using Recommender Systems to Support Continuous Ontology Development: Develops machine learning algorithms to assist users in collaboratively evolving ontologies by suggesting super-concepts for new terms.
6 Combating Attacks Against Social Tagging Environments: Analyzes the security vulnerabilities of social tagging systems by modeling attack strategies and evaluating their local and global impacts.
7 Conclusion: Summarizes the thesis findings, answers the core research questions, and provides an outlook on future research directions.
Keywords
Social Web, Recommender Systems, Social Tagging, Folksonomy, Ontology Learning, Machine Learning, Data Mining, Link Analysis, PageRank, Profile Injection Attacks, Information Retrieval, Ambiguity, Redundancy, Security, Personalization
Frequently Asked Questions
What is the core focus of this research?
This thesis examines the use of data mining and machine learning to improve social Web applications, specifically focusing on recommender systems for tagging environments and the security of these systems against malicious manipulation.
What are the primary themes addressed in the work?
The work covers four main areas: matching recommendation technology to specific domains, improving tag recommendation algorithms via link analysis, supporting collaborative ontology development, and analyzing security threats to social tagging.
What is the main objective of the proposed research?
The goal is to enhance the user experience in social tagging systems by simplifying navigation and contribution, while simultaneously increasing the robustness of the system against adversarial attacks.
Which scientific methods are primarily employed?
The thesis utilizes graph theory, link analysis algorithms (like PageRank), machine learning for classification tasks, and meta-analysis of existing recommendation literature.
What is covered in the main body of the thesis?
The main body details a framework for domain-specific recommendation, a weighted directed graph approach for tag recommendation, a machine learning approach to ontology development, and a systematic framework for analyzing and evaluating attacks in social tagging systems.
Which keywords best characterize the work?
The key concepts include Social Web, Social Tagging, Folksonomy, Recommender Systems, Link Analysis, and Profile Injection Attacks.
How does the proposed directed graph model improve folksonomy analysis?
Unlike undirected models, the directed graph accounts for non-symmetric information flow between users, tags, and resources, allowing for more precise importance weighting and link analysis.
What is the motivation behind studying attacks on social tagging systems?
Because social tagging systems are open to the public and rely on user-contributed content, they are highly susceptible to malicious profile injection. Understanding these attacks is essential to develop secure, robust, and reliable recommendation systems.
- Arbeit zitieren
- Maryam Ramezani (Autor:in), 2011, Using Data Mining for Facilitating User Contributions in the Social Semantic Web, München, GRIN Verlag, https://www.hausarbeiten.de/document/181536
