toc
Preface
Contents
Contributors
1 Recommender Systems: Introduction and Challenges
1.1 Introduction
1.2 Recommender Systems' Function
1.3 Data and Knowledge Sources
1.4 Recommendation Techniques
1.5 Recommender Systems Evaluation
1.6 Recommender Systems Applications
1.7 Recommender Systems and Human Computer Interaction
1.8 Advanced Topics
1.9 Challenges
1.9.1 Preference Acquisition and Profiling
1.9.2 Interaction
1.9.3 New Recommendation Tasks
References
Part I Recommendation Techniques
2 A Comprehensive Survey of Neighborhood-Based Recommendation Methods
2.1 Introduction
2.1.1 Advantages of Neighborhood Approaches
2.1.2 Objectives and Outline
2.2 Problem Definition and Notation
2.3 Neighborhood-Based Recommendation
2.3.1 User-Based Rating Prediction
2.3.2 User-Based Classification
2.3.3 Regression vs Classification
2.3.4 Item-Based Recommendation
2.3.5 User-Based vs Item-Based Recommendation
2.4 Components of Neighborhood Methods
2.4.1 Rating Normalization
2.4.1.1 Mean-Centering
2.4.1.2 Z-Score Normalization
2.4.1.3 Choosing a Normalization Scheme
2.4.2 Similarity Weight Computation
2.4.2.1 Correlation-Based Similarity
2.4.2.2 Other Similarity Measures
2.4.2.3 Considering the Significance of Weights
2.4.2.4 Considering the Variance of Ratings
2.4.2.5 Considering the Target Item
2.4.3 Neighborhood Selection
2.4.3.1 Pre-filtering of Neighbors
2.4.3.2 Neighbors in the Predictions
2.5 Advanced Techniques
2.5.1 Graph-Based Methods
2.5.1.1 Path-Based Similarity
2.5.1.2 Random Walk Similarity
2.5.2 Learning-Based Methods
2.5.2.1 Factorization Methods
2.5.2.2 Neighborhood-Learning Methods
2.6 Conclusion
References
3 Advances in Collaborative Filtering
3.1 Introduction
3.2 Preliminaries
3.2.1 Baseline Predictors
3.2.2 The Netflix Data
3.2.3 Implicit Feedback
3.3 Matrix Factorization Models
3.3.1 SVD
3.3.2 SVD++
3.3.3 Time-Aware Factor Model
3.3.3.1 Time Changing Baseline Predictors
3.3.3.2 Time Changing Factor Model
3.3.4 Comparison
3.3.4.1 Predicting Future Days
3.3.5 Summary
3.4 Neighborhood Models
3.4.1 Similarity Measures
3.4.2 Similarity-Based Interpolation
3.4.3 Jointly Derived Interpolation Weights
3.4.3.1 Formal Model
3.4.3.2 Computational Issues
3.4.4 Summary
3.5 Enriching Neighborhood Models
3.5.1 A Global Neighborhood Model
3.5.1.1 Building the Model
3.5.1.2 Parameter Estimation
3.5.1.3 Comparison of Accuracy
3.5.2 A Factorized Neighborhood Model
3.5.2.1 Factoring Item-Item Relationships
3.5.2.2 A User-User Model
3.5.3 Temporal Dynamics at Neighborhood Models
3.5.4 Summary
3.6 Between Neighborhood and Factorization
References
4 Semantics-Aware Content-Based Recommender Systems
4.1 Introduction
4.2 Overview of Content-Based Recommender Systems
4.2.1 Keyword-Based Vector Space Model
4.2.2 Methods for Learning User Profiles
4.2.2.1 Probabilistic Methods
4.2.2.2 Relevance Feedback
4.2.2.3 Nearest Neighbors
4.2.3 Advantages and Drawbacks of Content-Based Filtering
4.3 Top-Down Semantic Approaches
4.3.1 Approaches Based on Ontological Resources
4.3.2 Approaches Based on Unstructured or Semi-Structured Encyclopedic Knowledge
4.3.2.1 Explicit Semantic Analysis
4.3.2.2 CBRSs Leveraging Encyclopedic Knowledge
4.3.2.3 BabelNet: An Encyclopedic Dictionary
4.3.3 Approaches Based on Linked Open Data
4.3.3.1 CBRSs Leveraging Linked Open Data
4.3.3.2 (Other) Entity Linking Algorithms
4.4 Bottom-Up Semantic Approaches
4.4.1 Approaches Based on Discriminative Models
4.4.1.1 Dimensionality Reduction Techniques
4.4.1.2 Modeling Negation
4.4.1.3 CBRSs Leveraging Discriminative Models
4.5 Summary and Comparison of Approaches
4.6 Conclusions and Future Challenges
References
5 Constraint-Based Recommender Systems
5.1 Introduction
5.2 Development of Recommender Knowledge Bases
5.3 User Guidance in Recommendation Processes
5.4 Calculating Recommendations
5.5 Practical Experience from Fielded Applications
5.6 Future Research Issues
5.7 Summary
References
6 Context-Aware Recommender Systems
6.1 Introduction and Motivation
6.2 Context in Recommender Systems
6.2.1 What is Context?
6.2.2 Representational Approach to Modeling Contextual Information in Recommender Systems
6.2.3 Major Approaches to Modeling Contextual Information in Recommender Systems
6.2.4 Obtaining Contextual Information
6.3 Paradigms for Incorporating Representational Context in Recommender Systems
6.3.1 Contextual Pre-filtering
6.3.2 Contextual Post-filtering
6.3.3 Contextual Modeling
6.4 Discussion and Conclusions
References
7 Data Mining Methods for Recommender Systems
7.1 Introduction
7.2 Data Preprocessing
7.2.1 Similarity Measures
7.2.2 Sampling
7.2.3 Reducing Dimensionality
7.2.3.1 Principal Component Analysis
7.2.3.2 Matrix Factorization and Singular Value Decomposition
7.2.4 Denoising
7.3 Supervised Learning
7.3.1 Classification
7.3.1.1 Nearest Neighbors
7.3.1.2 Decision Trees
7.3.1.3 Ruled-Based Classifiers
7.3.1.4 Bayesian Classifiers
7.3.1.5 Logistic Regression
7.3.1.6 Support Vector Machines
7.3.1.7 Artificial Neural Networks
7.3.2 Ensembles of Classifiers
7.3.3 Evaluating Classifiers
7.4 Unsupervised Learning
7.4.1 Clustering
7.4.1.1 k-Means
7.4.1.2 Alternatives to k-Means
7.4.2 Association Rule Mining
7.5 Conclusions
References
Part II Recommender Systems Evaluation
8 Evaluating Recommender Systems
8.1 Introduction
8.2 Experimental Settings
8.2.1 Offline Experiments
8.2.1.1 Data Sets for Offline Experiments
8.2.1.2 Simulating User Behavior
8.2.1.3 More Complex User Modeling
8.2.2 User Studies
8.2.2.1 Advantages and Disadvantages
8.2.2.2 Between vs. Within Subjects
8.2.2.3 Variable Counter Balance
8.2.2.4 Questionnaires
8.2.3 Online Evaluation
8.2.4 Drawing Reliable Conclusions
8.2.4.1 Confidence and p-Values
8.2.4.2 Paired Results
8.2.4.3 Unpaired Results
8.2.4.4 Multiple Tests
8.2.4.5 Confidence Intervals
8.3 Recommender System Properties
8.3.1 User Preference
8.3.2 Prediction Accuracy
8.3.2.1 Measuring Ratings Prediction Accuracy
8.3.2.2 Measuring Usage Prediction
8.3.2.3 Ranking Measures
8.3.3 Coverage
8.3.3.1 Item Space Coverage
8.3.3.2 User Space Coverage
8.3.3.3 Cold-Start Problem
8.3.4 Confidence
8.3.5 Trust
8.3.6 Novelty
8.3.7 Serendipity
8.3.8 Diversity
8.3.9 Utility
8.3.10 Risk
8.3.11 Robustness
8.3.12 Privacy
8.3.13 Adaptivity
8.3.14 Scalability
8.4 Conclusion
References
9 Evaluating Recommender Systems with User Experiments
9.1 Introduction
9.2 Theoretical Foundation and Existing Work
9.2.1 Theoretical Foundation: The Knijnenburg et al. Evaluation Framework
9.2.2 Overview of Existing User-Centric Work and Promising Directions
9.2.2.1 Preference Elicitation Methods
9.2.2.2 Algorithms
9.2.2.3 Recommendations and Their Presentation
9.3 Practical Guidelines
9.3.1 Research Model
9.3.1.1 Determining Which OSAs Will Be Tested
9.3.1.2 Selecting Appropriate Outcome Measures (INT and EXP)
9.3.1.3 Explaining the Effects with Theory and Mediating Variables (SSAs)
9.3.1.4 Include PCs and SCs Where Appropriate
9.3.1.5 Practical Tip: Never Formulate a ``No Effect'' Hypothesis
9.3.2 Participants
9.3.2.1 Sampling Participants
9.3.2.2 Determining the Sample Size
9.3.2.3 Practical Tip: Run Your Studies on a Crowd-Sourcing Platform
9.3.3 Experimental Manipulations
9.3.3.1 Selecting Conditions to Test
9.3.3.2 Including Multiple Manipulations
9.3.3.3 Setting Up Between-Subjects or Within-Subjects Randomization
9.3.3.4 Practical Tip: Think Big, Start Small
9.3.4 Measurement
9.3.4.1 Creating Measurement Scales
9.3.4.2 Establishing Construct Validity
9.3.4.3 Practical Tip: Use Existing Scales
9.3.5 Statistical Evaluation
9.3.5.1 Piecewise Statistical Testing: T-tests, ANOVAs, and Regressions
9.3.5.2 Assumptions of Statistical Tests
9.3.5.3 Integrative Statistical Testing: Structural Equation Models
9.3.5.4 Practical Tip: Learn More About Structural Equation Modeling
9.4 Conclusion
References
10 Explaining Recommendations: Design and Evaluation
10.1 Introduction
10.2 Designing the Presentation and Interaction with Recommendations
10.2.1 Presenting Recommendations
10.2.2 Preference Elicitation
10.3 Explanation Styles
10.3.1 Collaborative-Based Style Explanations
10.3.2 Content-Based Style Explanation
10.3.3 Case-Based Reasoning (CBR) Style Explanations
10.3.4 Knowledge and Utility-Based Style Explanations
10.3.5 Demographic-Based Style Explanations
10.4 Goals and Metrics
10.4.1 Explain How the System Works: Transparency
10.4.2 Allow Users to Tell the System It Is Wrong: Scrutability
10.4.3 Increase Users' Confidence in the System: Trust
10.4.4 Convince Users to Try or Buy: Persuasiveness
10.4.5 Help Users Make Good Decisions: Effectiveness
10.4.6 Help Users Make Decisions Faster: Efficiency
10.4.7 Make the Use of the System Enjoyable: Satisfaction
10.5 Future Directions
10.5.1 Social Recommendations
10.5.2 Explanations, Serendipity and the Filter Bubble
10.5.3 When Should Explanations Be Shown?
10.5.4 Explanations: Help or Harm?
References
Part III Recommendation Techniques
11 Recommender Systems in Industry: A Netflix Case Study
11.1 Introduction
11.2 Recommender Systems in Industry
11.3 The Netflix Prize
11.3.1 Lessons from the Prize
11.4 Recommendation Beyond Rating Prediction
11.4.1 Everything Is a Recommendation
11.4.2 Ranking
11.4.3 Page Optimization
11.5 Data and Models
11.5.1 Data
11.5.2 Models
11.6 Consumer Data Science
11.7 Architectures
11.7.1 Event and Data Distribution
11.7.2 Offline, Nearline, and Online Computation
11.7.3 Recommendation Results
11.8 Research Directions with Industrial Applicability
11.8.1 Beyond Explicit Ratings
11.8.2 Personalized Learning to Rank
11.8.3 Full Page Optimization
11.8.4 Context-Aware Recommendations
11.8.5 Metrics and Evaluation
11.8.6 Class Imbalance Problems and Presentation Effects
11.8.7 Social Recommendations
11.9 Conclusion
References
12 Panorama of Recommender Systems to Support Learning
12.1 Introduction
12.2 Technology Enhanced Learning (TEL)
12.3 Classification Framework for TEL RecSys Review
12.4 Survey Results
12.4.1 Method and Overview of TEL RecSys
12.4.1.1 Cluster 1: Recommending Resources for Learning Based on Collaborative Filtering
12.4.1.2 Cluster 2: Improving Collaborative Filtering Algorithms with TEL Domain Particularities
12.4.1.3 Cluster 3: Educational Constraints as Source of Information for the Recommendation
Process
12.4.1.4 Cluster 4: Exploring Non Collaborative Filtering Techniques to Find Successful
Educational Recommendations
12.4.1.5 Cluster 5: Consider Contextual Information in the Recommendation Process
12.4.1.6 Cluster 6: Assessing the Educational Impact of Recommendations in Educational
Scenarios
12.4.1.7 Cluster 7: Recommending Courses
12.4.2 Analysis According to the Framework
12.5 Conclusions
References
13 Music Recommender Systems
13.1 Introduction
13.2 Content-Based Music Recommendation
13.2.1 Metadata Content
13.2.1.1 Manual Annotations
13.2.1.2 Social Tags
13.2.1.3 Annotations by Web Content Mining
13.2.2 Audio Content
13.2.2.1 Acoustic Features: Timbral, Temporal, and Tonal
13.2.2.2 Automatic Semantic Annotation
13.3 Contextual Music Recommendation
13.3.1 Environment-Related Context
13.3.2 User-Related Context
13.3.3 Incorporating Context Information in Music Recommender Systems
13.4 Hybrid Music Recommendation
13.4.1 Combining Content with Context Descriptors
13.4.2 Combining Collaborative Filtering with Content Descriptors
13.4.3 Combining Collaborative Filtering with Context Descriptors
13.5 Automatic Playlist Generation
13.5.1 Parallel and Serial Consumption
13.5.2 Playlist Evaluation
13.5.2.1 User Studies
13.5.2.2 Semantic Cohesion
13.5.2.3 Partial Playlist Prediction
13.5.2.4 Generative Likelihood
13.5.3 Playlist Generation Algorithms
13.5.3.1 Constraint Satisfaction
13.5.3.2 Similarity Heuristics
13.5.3.3 Machine Learning Approaches
13.6 Data Sets and Evaluation
13.6.1 Evaluation Methodologies
13.6.2 Yahoo! Music Dataset and KDD Cup 2011
13.6.3 Million Song Dataset (MSD) and MSD Challenge 2012
13.6.4 Last.fm Dataset: 360K/1K Users
13.6.5 MusicMicro and Million Musical Tweets Dataset (MMTD)
13.6.6 AotM-2011
13.7 Conclusions and Challenges
References
14 The Anatomy of Mobile Location-Based Recommender Systems
14.1 Introduction
14.1.1 Defining a Mobile Location-Based Recommender System
14.2 Data for Mobile Recommender Systems
14.2.1 Uncovering Points of Interest and Location Preferences
14.2.2 Behavioural Inferences from Smartphone Sensors
14.3 Computing Recommendations in Mobile Applications
14.3.1 Overview of Recommendation Formulations
14.3.2 Algorithmic Approaches to Venue Recommendation
14.4 Evaluating Mobile Recommendations
14.5 Conclusions and Future Directions
References
15 Social Recommender Systems
15.1 Introduction
15.2 Content Recommendation
15.2.1 Key Domains
15.2.2 Group Recommendation
15.2.3 Case Study: Social Media Recommendation in the Enterprise
15.2.4 Summary
15.3 People Recommendation
15.3.1 Recommending People to Connect With
15.3.2 Recommending Strangers
15.3.3 Recommending People to Follow
15.3.4 Related Research Areas
15.3.5 Summary
15.4 Discussion
15.5 Emerging Domains and Open Challenges
15.5.1 Emerging Domains
15.5.2 Open Challenges
References
16 People-to-People Reciprocal Recommenders
16.1 Introduction
16.2 Reciprocal vs Traditional Recommenders
16.3 Previous Work on People-to-People Recommenders
16.3.1 Social Networks
16.3.2 Mentor-Mentee Matching
16.3.3 Job Recommendation
16.3.4 Online Dating
16.4 A Case Study in Online Dating
16.4.1 A Content-Collaborative Reciprocal Recommender for Online Dating
16.4.1.1 Algorithm
16.4.1.2 Ranking Method Support
16.4.1.3 Evaluation
16.4.2 Explicit and Implicit User Preferences
16.4.2.1 Explicit User Preferences
16.4.2.2 Implicit User Preferences
16.4.2.3 Are Explicit Preferences Good Predictors of User Interactions?
16.4.2.4 Are Implicit Preferences Good Predictors of User Interactions?
16.4.2.5 Using User Preferences for Ranking Candidates in CCR
16.5 Conclusions and Future Work
References
17 Collaboration, Reputation and Recommender Systems in Social Web Search
17.1 Introduction
17.2 A Brief History of Web Search
17.3 The Future of Web Search
17.3.1 Personalizing Web Search
17.3.2 Collaborative Information Retrieval
17.3.3 On Reputation and Recommendation
17.3.4 Towards Social Search
17.4 Case-Study 1: HeyStaks—A Social Search Utility
17.4.1 The HeyStaks System
17.4.2 The HeyStaks Recommendation Engine
17.4.3 Evaluation
17.5 Case-Study 2: A Reputation Model for Social Search
17.5.1 From Activities to Reputation
17.5.2 Reputation as Collaboration
17.5.3 An Example
17.5.4 Graph-Based Reputation Models
17.5.4.1 Reputation as a Weighted Sum of Collaboration Events
17.5.4.2 Reputation as PageRank
17.5.5 From User Reputation to Result Promotion
17.5.5.1 Max Reputation
17.5.5.2 Hooper's Reputation
17.5.6 Evaluation
17.5.6.1 Dataset and Methodology
17.5.6.2 User Reputation
17.5.6.3 From Reputation to Quality
17.6 Search Futures
17.6.1 From Search to Discovery
17.6.2 Search in a Sensor-Rich, Mobile World
References
Part IV Human Computer Interaction
18 Human Decision Making and Recommender Systems
18.1 Introduction and Preview
18.2 Choice Patterns and Recommendation
18.2.1 Attribute-Based Choice
18.2.2 Consequence-Based Choice
18.2.3 Experience-Based Choice
18.2.4 Socially Based Choice
18.2.5 Policy-Based Choice
18.2.6 Trial-and-Error-Based Choice
18.2.7 Combinations of Choice Patterns
18.2.8 What Constitutes a Good Choice?
18.3 Choice Support Strategies and Recommendation
18.3.1 Evaluate on Behalf of the Chooser
18.3.2 Advise About Processing
18.3.3 Access Information and Experience
18.3.4 Represent the Choice Situation
18.3.5 Combine and Compute
18.3.6 Design the Domain
18.3.7 Concluding Remark on Support Strategies
18.4 Arguments and Explanations
18.4.1 Arguments
18.4.2 Explanations of Recommendations
18.4.2.1 Type 1: Direct Support for the Assessment of the Credibility of the Recommender
System
18.4.2.2 Type 2: An Argument Coupled with a Fidelity Claim
18.4.2.3 Type 3: An Explicit Description of the Recommender System's Processing
18.5 ``Preferences'' and Ratings
18.5.1 What Are ``Preferences''?
18.5.2 What Do Ratings Reflect?
18.5.2.1 A Sketch of the Processing Underlying Ratings
18.5.2.2 Implications for the Practice of Rating Elicitation
18.6 Combating Choice Overload
18.7 Supporting Trial and Error
18.7.1 Trial and Error with Stable Evaluation Criteria
18.7.2 Trial and Error with Evolving Evaluation Criteria
18.8 Dealing with Potentially Distorting Influences on Choice Processes
18.8.1 Context Effects
18.8.2 Order Effects
18.8.3 Framing Effects
18.8.4 Priming Effects
18.8.5 Defaults
18.9 Recapitulation and Concluding Remarks
References
19 Privacy Aspects of Recommender Systems
19.1 Introduction
19.2 Privacy Risks in Recommender Systems
19.2.1 Risks Imposed by the Recommender System
19.2.1.1 Direct Access to Data
19.2.1.2 Inference from User Preference Data
19.2.2 Risks Imposed by Other System Users
19.2.3 Risks Imposed by External Entities
19.2.4 Summary
19.3 Privacy Solutions
19.3.1 Architecture and System Design Solutions
19.3.1.1 Trusted Software for Limiting Linkability and Propagation of User Data
19.3.1.2 User-Managed Portable Profiles
19.3.1.3 Generating Recommendations on the Client
19.3.2 Algorithmic Solutions
19.3.2.1 Pseudonyms and Anonymization
19.3.2.2 Obfuscation
19.3.2.3 Differential Privacy
19.3.2.4 Cryptographic Solutions
19.3.3 Policy Solutions
19.4 Human Aspects and Perception of Privacy
19.4.1 The Limits of Transparency and Control
19.4.2 Privacy Nudges
19.4.3 Privacy Adaptation
19.5 Summary and Discussion
References
20 Source Factors in Recommender System Credibility Evaluation
20.1 Introduction
20.2 Credibility Evaluation of Online Sources
20.3 Recommender Systems as Social Actors
20.4 Source Factors in Human-Human Communication
20.4.1 Source Credibility
20.4.2 Source Cues
20.4.2.1 Source Likeability
20.4.2.2 Multiple Sources
20.4.2.3 Similarity
20.4.2.4 Symbols of Authority
20.4.2.5 Styles of Speech
20.4.2.6 Humor
20.4.2.7 Physical Attractiveness
20.4.2.8 Caring
20.4.2.9 Familiarity and Friendliness
20.4.2.10 Discussion
20.5 Source Factors in Human-Technology Interactions
20.6 Source Factors in Human-Recommender System Interactions
20.6.1 Recommender System Type
20.6.2 Input Characteristics
20.6.3 Process Characteristics
20.6.4 Output Characteristics
20.6.5 Characteristics of Embodied Agents
20.6.6 Impact of Emerging Social Technologies
20.7 Discussion
20.8 Implications
20.9 Directions for Future Research
References
21 Personality and Recommender Systems
21.1 Introduction
21.2 What is Personality?
21.2.1 The Five Factor Model of Personality
21.2.2 Other Models of Personality
21.2.3 How Does Personality Relate to User Preferences?
21.3 Personality Acquisition
21.3.1 Explicit Personality Acquisition
21.3.2 Implicit Personality Acquisition
21.3.3 Datasets for Offline Recommender SystemsExperiments
21.4 How to Use Personality in Recommender Systems
21.4.1 Addressing the New User Problem
21.4.2 Diversity/Serendipity
21.4.3 Cross-Domain Recommendations
21.4.4 Group Recommender Systems
21.5 Open Issues and Challenges
21.5.1 Non-intrusive Acquisition of Personality Information
21.5.2 Larger Datasets
21.5.3 Cross-Domain Applications
21.5.4 Diversity
21.5.5 Privacy Issues
21.6 Conclusion
References
Part V Advanced Topics
22 Group Recommender Systems: Aggregation, Satisfaction and Group Attributes
22.1 Introduction
22.2 Usage Scenarios and Classification of Group Recommenders
22.2.1 Usage Scenario 1: Interactive Television
22.2.2 Usage Scenario 2: Ambient Intelligence
22.2.3 Usage Scenarios Underlying Related Work
22.2.4 A Classification of Group Recommenders
22.3 Aggregation Strategies
22.3.1 Overview of Aggregation Strategies
22.3.2 Aggregation Strategies Used in Related Work
22.3.3 Which Strategy Performs The Best
22.4 Impact of Sequence Order
22.5 Modeling Affective State
22.5.1 Modeling an Individual's Satisfaction on Its Own
22.5.2 Effects of the Group on an Individual's Satisfaction
22.6 Using Satisfaction Inside Aggregation Strategies
22.7 Incorporating Group Attributes: Roles, Personality, Expertise, Relationship Strength, Relationship
Type and Personal Impact
22.8 Applying Group Recommendation to Individual Users
22.8.1 Multiple Criteria
22.8.2 Cold-Start Problem
22.8.3 Virtual Group Members
22.9 Conclusions and Challenges
22.9.1 Main Issues Raised
22.9.2 Caveat: Group Modeling
22.9.3 Challenges
References
23 Aggregation Functions for Recommender Systems
23.1 Introduction
23.2 Types of Aggregation in Recommender Systems
23.2.1 Aggregation of Preferences in CF
23.2.2 Aggregation of Features in CB and UB Recommendation
23.2.3 Item and User Similarity and Neighborhood Formation
23.2.4 Profile Construction for CB, UB
23.2.5 Connectives in Case-Based Reasoning for RS
23.2.6 Weighted Hybrid Systems
23.3 Review of Aggregation Functions
23.3.1 Definitions and Properties
23.3.1.1 Practical Considerations in RS
23.3.2 Aggregation Families
23.3.2.1 Quasi-Arithmetic Means
23.3.2.2 OWA Functions
23.3.2.3 Choquet and Sugeno Integrals
23.3.2.4 T-Norms and T-Conorms
23.3.2.5 Nullnorms and Uninorms
23.4 Construction of Aggregation Functions
23.4.1 Data Collection and Preprocessing
23.4.2 Desired Properties, Semantics and Interpretation
23.4.3 Complexity and the Understanding of Function Behavior
23.4.4 Penalty-Based Construction
23.4.5 Weight and Parameter Determination
23.5 Sophisticated Aggregation Procedures in Recommender Systems: Tailoring for Specific
Applications
23.6 Conclusions
References
24 Active Learning in Recommender Systems
24.1 Introduction
24.1.1 Objectives of Active Learning in Recommender Systems
24.1.2 An Illustrative Example
24.1.3 Types of Active Learning
24.2 Properties of Data Points
24.2.1 Other Considerations
24.3 Active Learning in Recommender Systems
24.3.1 Active Learning Formulation
24.4 Uncertainty-Based Active Learning
24.4.1 Output Uncertainty
24.4.1.1 Active Learning Methods
24.4.1.2 Uncertainty Measurement
24.4.2 Decision Boundary Uncertainty
24.4.3 Model Uncertainty
24.4.3.1 Probabilistic Models
24.5 Error-Based Active Learning
24.5.1 Instance-Based Methods
24.5.1.1 Output Estimates Change (Y-Change)
24.5.1.2 Cross Validation-Based
24.5.2 Model-Based
24.5.2.1 Parameter Change-Based
24.5.2.2 Variance-Based
24.5.2.3 Image Restoration-Based
24.6 Ensemble-Based Active Learning
24.6.1 Models-Based
24.6.2 Candidates-Based
24.7 Conversation-Based Active Learning
24.7.1 Case-Based Critique
24.7.2 Diversity-Based
24.7.3 Query Editing-Based
24.8 Evaluation Settings
24.8.1 Scope
24.8.2 Natural Rating Acquisition
24.8.3 Temporal Evolution
24.8.4 Ratability
24.8.5 Summary
24.9 Computational Considerations
24.10 Discussion
References
25 Multi-Criteria Recommender Systems
25.1 Introduction
25.2 Multi-Criteria Rating Recommendation
25.2.1 Traditional Single-Rating Recommendation Problem
25.2.2 Extending Traditional Recommender Systems to Include Multi-Criteria Ratings
25.3 Engaging Multi-Criteria Ratings During Prediction
25.3.1 Heuristic Approaches
25.3.2 Model-Based Approaches
25.4 Engaging Multi-Criteria Ratings During Recommendation
25.4.1 Related Work: Multi-Criteria Optimization
25.4.2 Designing a Total Order for Item Recommendations
25.4.3 Finding Pareto Optimal Item Recommendations
25.4.4 Using Multi-Criteria Ratings as Recommendation Filters
25.5 Discussion and Future Work
25.5.1 Developing New Approaches for Multi-Criteria Ratings
25.5.2 Extending Existing Techniques for Multi-Criteria Settings
25.5.3 Managing Multi-Criteria Ratings
25.6 Conclusions
References
26 Novelty and Diversity in Recommender Systems
26.1 Introduction
26.2 Novelty and Diversity in Recommender Systems
26.2.1 Why Novelty and Diversity in Recommendation
26.2.2 Defining Novelty and Diversity
26.2.3 Diversity in Other Fields
26.3 Novelty and Diversity Evaluation
26.3.1 Notation
26.3.2 Average Intra-List Distance
26.3.3 Global Long-Tail Novelty
26.3.4 User-Specific Unexpectedness
26.3.5 Inter-Recommendation Diversity Metrics
26.3.6 Specific Methodologies
26.3.7 Diversity vs. Novelty vs. Serendipity
26.3.8 Information Retrieval Diversity
26.4 Novelty and Diversity Enhancement Approaches
26.4.1 Result Diversification/Re-ranking
26.4.2 Using Clustering for Diversification
26.4.3 Fusion-Based Methods
26.4.4 Learning to Rank with Diversity
26.4.5 Serendipity: Enabling Surprising Recommendations
26.4.6 Other Approaches
26.4.7 User Studies
26.4.8 Diversification Approaches in Information Retrieval
26.5 Unified View
26.5.1 General Novelty/Diversity Metric Scheme
26.5.2 Item Novelty Models
26.5.2.1 Item Discovery
26.5.2.2 Item Familiarity
26.5.3 Resulting Metrics
26.5.3.1 Discovery-Based
26.5.3.2 Familiarity-Based
26.5.3.3 Further Unification
26.5.3.4 Direct Optimization of Novelty Models
26.5.4 Connecting Recommendation Diversity and Search Diversity
26.5.4.1 Rank and Relevance
26.5.4.2 IR Diversity in Recommendation
26.5.4.3 Personalized Diversity
26.6 Empirical Metric Comparison
26.7 Conclusion
References
27 Cross-Domain Recommender Systems
27.1 Introduction
27.2 Formulation of the Cross-Domain Recommendation Problem
27.2.1 Definition of Domain
27.2.2 Cross-Domain Recommendation Tasks
27.2.3 Cross-Domain Recommendation Goals
27.2.4 Cross-Domain Recommendation Scenarios
27.3 Categorization of Cross-Domain Recommendation Techniques
27.4 Knowledge Aggregation for Cross-Domain Recommendations
27.4.1 Merging Single-Domain User Preferences
27.4.2 Mediating Single-Domain User Modeling Data
27.4.3 Combining Single-Domain Recommendations
27.5 Knowledge Linkage and Transfer for Cross-Domain Recommendation
27.5.1 Linking Domains
27.5.2 Sharing Latent Features by Domains
27.5.3 Transferring Rating Patterns Between Domains
27.6 Evaluation of Cross-Domain Recommender Systems
27.6.1 Data Partitioning
27.6.2 Metrics
27.6.3 Sensitivity Analysis
27.7 Practical Considerations in Cross-Domain Recommendation
27.8 Open Research Issues
References
28 Robust Collaborative Recommendation
28.1 Introduction
28.2 Defining the Problem
28.2.1 An Example Attack
28.3 Characterising Attacks
28.3.1 Basic Attacks
28.3.1.1 Random Attack
28.3.1.2 Average Attack
28.3.2 Low-Knowledge Attacks
28.3.2.1 Bandwagon Attack
28.3.2.2 Segment Attack
28.3.3 Nuke Attack Models
28.3.3.1 Love/Hate Attack
28.3.3.2 Reverse Bandwagon Attack
28.3.4 Informed Attack Models
28.3.4.1 Popular Attack
28.3.4.2 Probe Attack Strategy
28.3.4.3 Power User Attack
28.3.5 Obfuscated Attacks
28.4 Measuring Robustness
28.4.1 Evaluation Metrics
28.4.2 Push Attacks
28.4.3 Nuke Attacks
28.4.4 Informed Attacks
28.4.5 Attack Impact
28.5 Attack Detection
28.5.1 Evaluation Metrics
28.5.1.1 Impact on Recommender and Attack Performance
28.5.2 Single Profile Detection
28.5.2.1 Unsupervised Detection
28.5.2.2 Supervised Detection
28.5.3 Group Profile Detection
28.5.3.1 Neighbourhood Filtering
28.5.3.2 Detecting Attacks Using Profile Clustering
28.5.3.3 Hybrid Attack Detection
28.5.4 Detection Findings
28.6 Beyond Memory-Based Algorithms
28.6.1 Model-Based Recommendation
28.6.2 Privacy-Preserving Algorithms
28.6.3 The Influence Limiter and Trust-Based Recommendation
28.7 Robust Algorithms
28.7.1 Robust Matrix Factorisation (RMF)
28.7.2 Other Robust Recommendation Algorithms
28.8 Practical Countermeasures to Recommender System Attack
28.9 Conclusion
References
Index
Contents
Contributors
1 Recommender Systems: Introduction and Challenges
1.1 Introduction
1.2 Recommender Systems' Function
1.3 Data and Knowledge Sources
1.4 Recommendation Techniques
1.5 Recommender Systems Evaluation
1.6 Recommender Systems Applications
1.7 Recommender Systems and Human Computer Interaction
1.8 Advanced Topics
1.9 Challenges
1.9.1 Preference Acquisition and Profiling
1.9.2 Interaction
1.9.3 New Recommendation Tasks
References
Part I Recommendation Techniques
2 A Comprehensive Survey of Neighborhood-Based Recommendation Methods
2.1 Introduction
2.1.1 Advantages of Neighborhood Approaches
2.1.2 Objectives and Outline
2.2 Problem Definition and Notation
2.3 Neighborhood-Based Recommendation
2.3.1 User-Based Rating Prediction
2.3.2 User-Based Classification
2.3.3 Regression vs Classification
2.3.4 Item-Based Recommendation
2.3.5 User-Based vs Item-Based Recommendation
2.4 Components of Neighborhood Methods
2.4.1 Rating Normalization
2.4.1.1 Mean-Centering
2.4.1.2 Z-Score Normalization
2.4.1.3 Choosing a Normalization Scheme
2.4.2 Similarity Weight Computation
2.4.2.1 Correlation-Based Similarity
2.4.2.2 Other Similarity Measures
2.4.2.3 Considering the Significance of Weights
2.4.2.4 Considering the Variance of Ratings
2.4.2.5 Considering the Target Item
2.4.3 Neighborhood Selection
2.4.3.1 Pre-filtering of Neighbors
2.4.3.2 Neighbors in the Predictions
2.5 Advanced Techniques
2.5.1 Graph-Based Methods
2.5.1.1 Path-Based Similarity
2.5.1.2 Random Walk Similarity
2.5.2 Learning-Based Methods
2.5.2.1 Factorization Methods
2.5.2.2 Neighborhood-Learning Methods
2.6 Conclusion
References
3 Advances in Collaborative Filtering
3.1 Introduction
3.2 Preliminaries
3.2.1 Baseline Predictors
3.2.2 The Netflix Data
3.2.3 Implicit Feedback
3.3 Matrix Factorization Models
3.3.1 SVD
3.3.2 SVD++
3.3.3 Time-Aware Factor Model
3.3.3.1 Time Changing Baseline Predictors
3.3.3.2 Time Changing Factor Model
3.3.4 Comparison
3.3.4.1 Predicting Future Days
3.3.5 Summary
3.4 Neighborhood Models
3.4.1 Similarity Measures
3.4.2 Similarity-Based Interpolation
3.4.3 Jointly Derived Interpolation Weights
3.4.3.1 Formal Model
3.4.3.2 Computational Issues
3.4.4 Summary
3.5 Enriching Neighborhood Models
3.5.1 A Global Neighborhood Model
3.5.1.1 Building the Model
3.5.1.2 Parameter Estimation
3.5.1.3 Comparison of Accuracy
3.5.2 A Factorized Neighborhood Model
3.5.2.1 Factoring Item-Item Relationships
3.5.2.2 A User-User Model
3.5.3 Temporal Dynamics at Neighborhood Models
3.5.4 Summary
3.6 Between Neighborhood and Factorization
References
4 Semantics-Aware Content-Based Recommender Systems
4.1 Introduction
4.2 Overview of Content-Based Recommender Systems
4.2.1 Keyword-Based Vector Space Model
4.2.2 Methods for Learning User Profiles
4.2.2.1 Probabilistic Methods
4.2.2.2 Relevance Feedback
4.2.2.3 Nearest Neighbors
4.2.3 Advantages and Drawbacks of Content-Based Filtering
4.3 Top-Down Semantic Approaches
4.3.1 Approaches Based on Ontological Resources
4.3.2 Approaches Based on Unstructured or Semi-Structured Encyclopedic Knowledge
4.3.2.1 Explicit Semantic Analysis
4.3.2.2 CBRSs Leveraging Encyclopedic Knowledge
4.3.2.3 BabelNet: An Encyclopedic Dictionary
4.3.3 Approaches Based on Linked Open Data
4.3.3.1 CBRSs Leveraging Linked Open Data
4.3.3.2 (Other) Entity Linking Algorithms
4.4 Bottom-Up Semantic Approaches
4.4.1 Approaches Based on Discriminative Models
4.4.1.1 Dimensionality Reduction Techniques
4.4.1.2 Modeling Negation
4.4.1.3 CBRSs Leveraging Discriminative Models
4.5 Summary and Comparison of Approaches
4.6 Conclusions and Future Challenges
References
5 Constraint-Based Recommender Systems
5.1 Introduction
5.2 Development of Recommender Knowledge Bases
5.3 User Guidance in Recommendation Processes
5.4 Calculating Recommendations
5.5 Practical Experience from Fielded Applications
5.6 Future Research Issues
5.7 Summary
References
6 Context-Aware Recommender Systems
6.1 Introduction and Motivation
6.2 Context in Recommender Systems
6.2.1 What is Context?
6.2.2 Representational Approach to Modeling Contextual Information in Recommender Systems
6.2.3 Major Approaches to Modeling Contextual Information in Recommender Systems
6.2.4 Obtaining Contextual Information
6.3 Paradigms for Incorporating Representational Context in Recommender Systems
6.3.1 Contextual Pre-filtering
6.3.2 Contextual Post-filtering
6.3.3 Contextual Modeling
6.4 Discussion and Conclusions
References
7 Data Mining Methods for Recommender Systems
7.1 Introduction
7.2 Data Preprocessing
7.2.1 Similarity Measures
7.2.2 Sampling
7.2.3 Reducing Dimensionality
7.2.3.1 Principal Component Analysis
7.2.3.2 Matrix Factorization and Singular Value Decomposition
7.2.4 Denoising
7.3 Supervised Learning
7.3.1 Classification
7.3.1.1 Nearest Neighbors
7.3.1.2 Decision Trees
7.3.1.3 Ruled-Based Classifiers
7.3.1.4 Bayesian Classifiers
7.3.1.5 Logistic Regression
7.3.1.6 Support Vector Machines
7.3.1.7 Artificial Neural Networks
7.3.2 Ensembles of Classifiers
7.3.3 Evaluating Classifiers
7.4 Unsupervised Learning
7.4.1 Clustering
7.4.1.1 k-Means
7.4.1.2 Alternatives to k-Means
7.4.2 Association Rule Mining
7.5 Conclusions
References
Part II Recommender Systems Evaluation
8 Evaluating Recommender Systems
8.1 Introduction
8.2 Experimental Settings
8.2.1 Offline Experiments
8.2.1.1 Data Sets for Offline Experiments
8.2.1.2 Simulating User Behavior
8.2.1.3 More Complex User Modeling
8.2.2 User Studies
8.2.2.1 Advantages and Disadvantages
8.2.2.2 Between vs. Within Subjects
8.2.2.3 Variable Counter Balance
8.2.2.4 Questionnaires
8.2.3 Online Evaluation
8.2.4 Drawing Reliable Conclusions
8.2.4.1 Confidence and p-Values
8.2.4.2 Paired Results
8.2.4.3 Unpaired Results
8.2.4.4 Multiple Tests
8.2.4.5 Confidence Intervals
8.3 Recommender System Properties
8.3.1 User Preference
8.3.2 Prediction Accuracy
8.3.2.1 Measuring Ratings Prediction Accuracy
8.3.2.2 Measuring Usage Prediction
8.3.2.3 Ranking Measures
8.3.3 Coverage
8.3.3.1 Item Space Coverage
8.3.3.2 User Space Coverage
8.3.3.3 Cold-Start Problem
8.3.4 Confidence
8.3.5 Trust
8.3.6 Novelty
8.3.7 Serendipity
8.3.8 Diversity
8.3.9 Utility
8.3.10 Risk
8.3.11 Robustness
8.3.12 Privacy
8.3.13 Adaptivity
8.3.14 Scalability
8.4 Conclusion
References
9 Evaluating Recommender Systems with User Experiments
9.1 Introduction
9.2 Theoretical Foundation and Existing Work
9.2.1 Theoretical Foundation: The Knijnenburg et al. Evaluation Framework
9.2.2 Overview of Existing User-Centric Work and Promising Directions
9.2.2.1 Preference Elicitation Methods
9.2.2.2 Algorithms
9.2.2.3 Recommendations and Their Presentation
9.3 Practical Guidelines
9.3.1 Research Model
9.3.1.1 Determining Which OSAs Will Be Tested
9.3.1.2 Selecting Appropriate Outcome Measures (INT and EXP)
9.3.1.3 Explaining the Effects with Theory and Mediating Variables (SSAs)
9.3.1.4 Include PCs and SCs Where Appropriate
9.3.1.5 Practical Tip: Never Formulate a ``No Effect'' Hypothesis
9.3.2 Participants
9.3.2.1 Sampling Participants
9.3.2.2 Determining the Sample Size
9.3.2.3 Practical Tip: Run Your Studies on a Crowd-Sourcing Platform
9.3.3 Experimental Manipulations
9.3.3.1 Selecting Conditions to Test
9.3.3.2 Including Multiple Manipulations
9.3.3.3 Setting Up Between-Subjects or Within-Subjects Randomization
9.3.3.4 Practical Tip: Think Big, Start Small
9.3.4 Measurement
9.3.4.1 Creating Measurement Scales
9.3.4.2 Establishing Construct Validity
9.3.4.3 Practical Tip: Use Existing Scales
9.3.5 Statistical Evaluation
9.3.5.1 Piecewise Statistical Testing: T-tests, ANOVAs, and Regressions
9.3.5.2 Assumptions of Statistical Tests
9.3.5.3 Integrative Statistical Testing: Structural Equation Models
9.3.5.4 Practical Tip: Learn More About Structural Equation Modeling
9.4 Conclusion
References
10 Explaining Recommendations: Design and Evaluation
10.1 Introduction
10.2 Designing the Presentation and Interaction with Recommendations
10.2.1 Presenting Recommendations
10.2.2 Preference Elicitation
10.3 Explanation Styles
10.3.1 Collaborative-Based Style Explanations
10.3.2 Content-Based Style Explanation
10.3.3 Case-Based Reasoning (CBR) Style Explanations
10.3.4 Knowledge and Utility-Based Style Explanations
10.3.5 Demographic-Based Style Explanations
10.4 Goals and Metrics
10.4.1 Explain How the System Works: Transparency
10.4.2 Allow Users to Tell the System It Is Wrong: Scrutability
10.4.3 Increase Users' Confidence in the System: Trust
10.4.4 Convince Users to Try or Buy: Persuasiveness
10.4.5 Help Users Make Good Decisions: Effectiveness
10.4.6 Help Users Make Decisions Faster: Efficiency
10.4.7 Make the Use of the System Enjoyable: Satisfaction
10.5 Future Directions
10.5.1 Social Recommendations
10.5.2 Explanations, Serendipity and the Filter Bubble
10.5.3 When Should Explanations Be Shown?
10.5.4 Explanations: Help or Harm?
References
Part III Recommendation Techniques
11 Recommender Systems in Industry: A Netflix Case Study
11.1 Introduction
11.2 Recommender Systems in Industry
11.3 The Netflix Prize
11.3.1 Lessons from the Prize
11.4 Recommendation Beyond Rating Prediction
11.4.1 Everything Is a Recommendation
11.4.2 Ranking
11.4.3 Page Optimization
11.5 Data and Models
11.5.1 Data
11.5.2 Models
11.6 Consumer Data Science
11.7 Architectures
11.7.1 Event and Data Distribution
11.7.2 Offline, Nearline, and Online Computation
11.7.3 Recommendation Results
11.8 Research Directions with Industrial Applicability
11.8.1 Beyond Explicit Ratings
11.8.2 Personalized Learning to Rank
11.8.3 Full Page Optimization
11.8.4 Context-Aware Recommendations
11.8.5 Metrics and Evaluation
11.8.6 Class Imbalance Problems and Presentation Effects
11.8.7 Social Recommendations
11.9 Conclusion
References
12 Panorama of Recommender Systems to Support Learning
12.1 Introduction
12.2 Technology Enhanced Learning (TEL)
12.3 Classification Framework for TEL RecSys Review
12.4 Survey Results
12.4.1 Method and Overview of TEL RecSys
12.4.1.1 Cluster 1: Recommending Resources for Learning Based on Collaborative Filtering
12.4.1.2 Cluster 2: Improving Collaborative Filtering Algorithms with TEL Domain Particularities
12.4.1.3 Cluster 3: Educational Constraints as Source of Information for the Recommendation
Process
12.4.1.4 Cluster 4: Exploring Non Collaborative Filtering Techniques to Find Successful
Educational Recommendations
12.4.1.5 Cluster 5: Consider Contextual Information in the Recommendation Process
12.4.1.6 Cluster 6: Assessing the Educational Impact of Recommendations in Educational
Scenarios
12.4.1.7 Cluster 7: Recommending Courses
12.4.2 Analysis According to the Framework
12.5 Conclusions
References
13 Music Recommender Systems
13.1 Introduction
13.2 Content-Based Music Recommendation
13.2.1 Metadata Content
13.2.1.1 Manual Annotations
13.2.1.2 Social Tags
13.2.1.3 Annotations by Web Content Mining
13.2.2 Audio Content
13.2.2.1 Acoustic Features: Timbral, Temporal, and Tonal
13.2.2.2 Automatic Semantic Annotation
13.3 Contextual Music Recommendation
13.3.1 Environment-Related Context
13.3.2 User-Related Context
13.3.3 Incorporating Context Information in Music Recommender Systems
13.4 Hybrid Music Recommendation
13.4.1 Combining Content with Context Descriptors
13.4.2 Combining Collaborative Filtering with Content Descriptors
13.4.3 Combining Collaborative Filtering with Context Descriptors
13.5 Automatic Playlist Generation
13.5.1 Parallel and Serial Consumption
13.5.2 Playlist Evaluation
13.5.2.1 User Studies
13.5.2.2 Semantic Cohesion
13.5.2.3 Partial Playlist Prediction
13.5.2.4 Generative Likelihood
13.5.3 Playlist Generation Algorithms
13.5.3.1 Constraint Satisfaction
13.5.3.2 Similarity Heuristics
13.5.3.3 Machine Learning Approaches
13.6 Data Sets and Evaluation
13.6.1 Evaluation Methodologies
13.6.2 Yahoo! Music Dataset and KDD Cup 2011
13.6.3 Million Song Dataset (MSD) and MSD Challenge 2012
13.6.4 Last.fm Dataset: 360K/1K Users
13.6.5 MusicMicro and Million Musical Tweets Dataset (MMTD)
13.6.6 AotM-2011
13.7 Conclusions and Challenges
References
14 The Anatomy of Mobile Location-Based Recommender Systems
14.1 Introduction
14.1.1 Defining a Mobile Location-Based Recommender System
14.2 Data for Mobile Recommender Systems
14.2.1 Uncovering Points of Interest and Location Preferences
14.2.2 Behavioural Inferences from Smartphone Sensors
14.3 Computing Recommendations in Mobile Applications
14.3.1 Overview of Recommendation Formulations
14.3.2 Algorithmic Approaches to Venue Recommendation
14.4 Evaluating Mobile Recommendations
14.5 Conclusions and Future Directions
References
15 Social Recommender Systems
15.1 Introduction
15.2 Content Recommendation
15.2.1 Key Domains
15.2.2 Group Recommendation
15.2.3 Case Study: Social Media Recommendation in the Enterprise
15.2.4 Summary
15.3 People Recommendation
15.3.1 Recommending People to Connect With
15.3.2 Recommending Strangers
15.3.3 Recommending People to Follow
15.3.4 Related Research Areas
15.3.5 Summary
15.4 Discussion
15.5 Emerging Domains and Open Challenges
15.5.1 Emerging Domains
15.5.2 Open Challenges
References
16 People-to-People Reciprocal Recommenders
16.1 Introduction
16.2 Reciprocal vs Traditional Recommenders
16.3 Previous Work on People-to-People Recommenders
16.3.1 Social Networks
16.3.2 Mentor-Mentee Matching
16.3.3 Job Recommendation
16.3.4 Online Dating
16.4 A Case Study in Online Dating
16.4.1 A Content-Collaborative Reciprocal Recommender for Online Dating
16.4.1.1 Algorithm
16.4.1.2 Ranking Method Support
16.4.1.3 Evaluation
16.4.2 Explicit and Implicit User Preferences
16.4.2.1 Explicit User Preferences
16.4.2.2 Implicit User Preferences
16.4.2.3 Are Explicit Preferences Good Predictors of User Interactions?
16.4.2.4 Are Implicit Preferences Good Predictors of User Interactions?
16.4.2.5 Using User Preferences for Ranking Candidates in CCR
16.5 Conclusions and Future Work
References
17 Collaboration, Reputation and Recommender Systems in Social Web Search
17.1 Introduction
17.2 A Brief History of Web Search
17.3 The Future of Web Search
17.3.1 Personalizing Web Search
17.3.2 Collaborative Information Retrieval
17.3.3 On Reputation and Recommendation
17.3.4 Towards Social Search
17.4 Case-Study 1: HeyStaks—A Social Search Utility
17.4.1 The HeyStaks System
17.4.2 The HeyStaks Recommendation Engine
17.4.3 Evaluation
17.5 Case-Study 2: A Reputation Model for Social Search
17.5.1 From Activities to Reputation
17.5.2 Reputation as Collaboration
17.5.3 An Example
17.5.4 Graph-Based Reputation Models
17.5.4.1 Reputation as a Weighted Sum of Collaboration Events
17.5.4.2 Reputation as PageRank
17.5.5 From User Reputation to Result Promotion
17.5.5.1 Max Reputation
17.5.5.2 Hooper's Reputation
17.5.6 Evaluation
17.5.6.1 Dataset and Methodology
17.5.6.2 User Reputation
17.5.6.3 From Reputation to Quality
17.6 Search Futures
17.6.1 From Search to Discovery
17.6.2 Search in a Sensor-Rich, Mobile World
References
Part IV Human Computer Interaction
18 Human Decision Making and Recommender Systems
18.1 Introduction and Preview
18.2 Choice Patterns and Recommendation
18.2.1 Attribute-Based Choice
18.2.2 Consequence-Based Choice
18.2.3 Experience-Based Choice
18.2.4 Socially Based Choice
18.2.5 Policy-Based Choice
18.2.6 Trial-and-Error-Based Choice
18.2.7 Combinations of Choice Patterns
18.2.8 What Constitutes a Good Choice?
18.3 Choice Support Strategies and Recommendation
18.3.1 Evaluate on Behalf of the Chooser
18.3.2 Advise About Processing
18.3.3 Access Information and Experience
18.3.4 Represent the Choice Situation
18.3.5 Combine and Compute
18.3.6 Design the Domain
18.3.7 Concluding Remark on Support Strategies
18.4 Arguments and Explanations
18.4.1 Arguments
18.4.2 Explanations of Recommendations
18.4.2.1 Type 1: Direct Support for the Assessment of the Credibility of the Recommender
System
18.4.2.2 Type 2: An Argument Coupled with a Fidelity Claim
18.4.2.3 Type 3: An Explicit Description of the Recommender System's Processing
18.5 ``Preferences'' and Ratings
18.5.1 What Are ``Preferences''?
18.5.2 What Do Ratings Reflect?
18.5.2.1 A Sketch of the Processing Underlying Ratings
18.5.2.2 Implications for the Practice of Rating Elicitation
18.6 Combating Choice Overload
18.7 Supporting Trial and Error
18.7.1 Trial and Error with Stable Evaluation Criteria
18.7.2 Trial and Error with Evolving Evaluation Criteria
18.8 Dealing with Potentially Distorting Influences on Choice Processes
18.8.1 Context Effects
18.8.2 Order Effects
18.8.3 Framing Effects
18.8.4 Priming Effects
18.8.5 Defaults
18.9 Recapitulation and Concluding Remarks
References
19 Privacy Aspects of Recommender Systems
19.1 Introduction
19.2 Privacy Risks in Recommender Systems
19.2.1 Risks Imposed by the Recommender System
19.2.1.1 Direct Access to Data
19.2.1.2 Inference from User Preference Data
19.2.2 Risks Imposed by Other System Users
19.2.3 Risks Imposed by External Entities
19.2.4 Summary
19.3 Privacy Solutions
19.3.1 Architecture and System Design Solutions
19.3.1.1 Trusted Software for Limiting Linkability and Propagation of User Data
19.3.1.2 User-Managed Portable Profiles
19.3.1.3 Generating Recommendations on the Client
19.3.2 Algorithmic Solutions
19.3.2.1 Pseudonyms and Anonymization
19.3.2.2 Obfuscation
19.3.2.3 Differential Privacy
19.3.2.4 Cryptographic Solutions
19.3.3 Policy Solutions
19.4 Human Aspects and Perception of Privacy
19.4.1 The Limits of Transparency and Control
19.4.2 Privacy Nudges
19.4.3 Privacy Adaptation
19.5 Summary and Discussion
References
20 Source Factors in Recommender System Credibility Evaluation
20.1 Introduction
20.2 Credibility Evaluation of Online Sources
20.3 Recommender Systems as Social Actors
20.4 Source Factors in Human-Human Communication
20.4.1 Source Credibility
20.4.2 Source Cues
20.4.2.1 Source Likeability
20.4.2.2 Multiple Sources
20.4.2.3 Similarity
20.4.2.4 Symbols of Authority
20.4.2.5 Styles of Speech
20.4.2.6 Humor
20.4.2.7 Physical Attractiveness
20.4.2.8 Caring
20.4.2.9 Familiarity and Friendliness
20.4.2.10 Discussion
20.5 Source Factors in Human-Technology Interactions
20.6 Source Factors in Human-Recommender System Interactions
20.6.1 Recommender System Type
20.6.2 Input Characteristics
20.6.3 Process Characteristics
20.6.4 Output Characteristics
20.6.5 Characteristics of Embodied Agents
20.6.6 Impact of Emerging Social Technologies
20.7 Discussion
20.8 Implications
20.9 Directions for Future Research
References
21 Personality and Recommender Systems
21.1 Introduction
21.2 What is Personality?
21.2.1 The Five Factor Model of Personality
21.2.2 Other Models of Personality
21.2.3 How Does Personality Relate to User Preferences?
21.3 Personality Acquisition
21.3.1 Explicit Personality Acquisition
21.3.2 Implicit Personality Acquisition
21.3.3 Datasets for Offline Recommender SystemsExperiments
21.4 How to Use Personality in Recommender Systems
21.4.1 Addressing the New User Problem
21.4.2 Diversity/Serendipity
21.4.3 Cross-Domain Recommendations
21.4.4 Group Recommender Systems
21.5 Open Issues and Challenges
21.5.1 Non-intrusive Acquisition of Personality Information
21.5.2 Larger Datasets
21.5.3 Cross-Domain Applications
21.5.4 Diversity
21.5.5 Privacy Issues
21.6 Conclusion
References
Part V Advanced Topics
22 Group Recommender Systems: Aggregation, Satisfaction and Group Attributes
22.1 Introduction
22.2 Usage Scenarios and Classification of Group Recommenders
22.2.1 Usage Scenario 1: Interactive Television
22.2.2 Usage Scenario 2: Ambient Intelligence
22.2.3 Usage Scenarios Underlying Related Work
22.2.4 A Classification of Group Recommenders
22.3 Aggregation Strategies
22.3.1 Overview of Aggregation Strategies
22.3.2 Aggregation Strategies Used in Related Work
22.3.3 Which Strategy Performs The Best
22.4 Impact of Sequence Order
22.5 Modeling Affective State
22.5.1 Modeling an Individual's Satisfaction on Its Own
22.5.2 Effects of the Group on an Individual's Satisfaction
22.6 Using Satisfaction Inside Aggregation Strategies
22.7 Incorporating Group Attributes: Roles, Personality, Expertise, Relationship Strength, Relationship
Type and Personal Impact
22.8 Applying Group Recommendation to Individual Users
22.8.1 Multiple Criteria
22.8.2 Cold-Start Problem
22.8.3 Virtual Group Members
22.9 Conclusions and Challenges
22.9.1 Main Issues Raised
22.9.2 Caveat: Group Modeling
22.9.3 Challenges
References
23 Aggregation Functions for Recommender Systems
23.1 Introduction
23.2 Types of Aggregation in Recommender Systems
23.2.1 Aggregation of Preferences in CF
23.2.2 Aggregation of Features in CB and UB Recommendation
23.2.3 Item and User Similarity and Neighborhood Formation
23.2.4 Profile Construction for CB, UB
23.2.5 Connectives in Case-Based Reasoning for RS
23.2.6 Weighted Hybrid Systems
23.3 Review of Aggregation Functions
23.3.1 Definitions and Properties
23.3.1.1 Practical Considerations in RS
23.3.2 Aggregation Families
23.3.2.1 Quasi-Arithmetic Means
23.3.2.2 OWA Functions
23.3.2.3 Choquet and Sugeno Integrals
23.3.2.4 T-Norms and T-Conorms
23.3.2.5 Nullnorms and Uninorms
23.4 Construction of Aggregation Functions
23.4.1 Data Collection and Preprocessing
23.4.2 Desired Properties, Semantics and Interpretation
23.4.3 Complexity and the Understanding of Function Behavior
23.4.4 Penalty-Based Construction
23.4.5 Weight and Parameter Determination
23.5 Sophisticated Aggregation Procedures in Recommender Systems: Tailoring for Specific
Applications
23.6 Conclusions
References
24 Active Learning in Recommender Systems
24.1 Introduction
24.1.1 Objectives of Active Learning in Recommender Systems
24.1.2 An Illustrative Example
24.1.3 Types of Active Learning
24.2 Properties of Data Points
24.2.1 Other Considerations
24.3 Active Learning in Recommender Systems
24.3.1 Active Learning Formulation
24.4 Uncertainty-Based Active Learning
24.4.1 Output Uncertainty
24.4.1.1 Active Learning Methods
24.4.1.2 Uncertainty Measurement
24.4.2 Decision Boundary Uncertainty
24.4.3 Model Uncertainty
24.4.3.1 Probabilistic Models
24.5 Error-Based Active Learning
24.5.1 Instance-Based Methods
24.5.1.1 Output Estimates Change (Y-Change)
24.5.1.2 Cross Validation-Based
24.5.2 Model-Based
24.5.2.1 Parameter Change-Based
24.5.2.2 Variance-Based
24.5.2.3 Image Restoration-Based
24.6 Ensemble-Based Active Learning
24.6.1 Models-Based
24.6.2 Candidates-Based
24.7 Conversation-Based Active Learning
24.7.1 Case-Based Critique
24.7.2 Diversity-Based
24.7.3 Query Editing-Based
24.8 Evaluation Settings
24.8.1 Scope
24.8.2 Natural Rating Acquisition
24.8.3 Temporal Evolution
24.8.4 Ratability
24.8.5 Summary
24.9 Computational Considerations
24.10 Discussion
References
25 Multi-Criteria Recommender Systems
25.1 Introduction
25.2 Multi-Criteria Rating Recommendation
25.2.1 Traditional Single-Rating Recommendation Problem
25.2.2 Extending Traditional Recommender Systems to Include Multi-Criteria Ratings
25.3 Engaging Multi-Criteria Ratings During Prediction
25.3.1 Heuristic Approaches
25.3.2 Model-Based Approaches
25.4 Engaging Multi-Criteria Ratings During Recommendation
25.4.1 Related Work: Multi-Criteria Optimization
25.4.2 Designing a Total Order for Item Recommendations
25.4.3 Finding Pareto Optimal Item Recommendations
25.4.4 Using Multi-Criteria Ratings as Recommendation Filters
25.5 Discussion and Future Work
25.5.1 Developing New Approaches for Multi-Criteria Ratings
25.5.2 Extending Existing Techniques for Multi-Criteria Settings
25.5.3 Managing Multi-Criteria Ratings
25.6 Conclusions
References
26 Novelty and Diversity in Recommender Systems
26.1 Introduction
26.2 Novelty and Diversity in Recommender Systems
26.2.1 Why Novelty and Diversity in Recommendation
26.2.2 Defining Novelty and Diversity
26.2.3 Diversity in Other Fields
26.3 Novelty and Diversity Evaluation
26.3.1 Notation
26.3.2 Average Intra-List Distance
26.3.3 Global Long-Tail Novelty
26.3.4 User-Specific Unexpectedness
26.3.5 Inter-Recommendation Diversity Metrics
26.3.6 Specific Methodologies
26.3.7 Diversity vs. Novelty vs. Serendipity
26.3.8 Information Retrieval Diversity
26.4 Novelty and Diversity Enhancement Approaches
26.4.1 Result Diversification/Re-ranking
26.4.2 Using Clustering for Diversification
26.4.3 Fusion-Based Methods
26.4.4 Learning to Rank with Diversity
26.4.5 Serendipity: Enabling Surprising Recommendations
26.4.6 Other Approaches
26.4.7 User Studies
26.4.8 Diversification Approaches in Information Retrieval
26.5 Unified View
26.5.1 General Novelty/Diversity Metric Scheme
26.5.2 Item Novelty Models
26.5.2.1 Item Discovery
26.5.2.2 Item Familiarity
26.5.3 Resulting Metrics
26.5.3.1 Discovery-Based
26.5.3.2 Familiarity-Based
26.5.3.3 Further Unification
26.5.3.4 Direct Optimization of Novelty Models
26.5.4 Connecting Recommendation Diversity and Search Diversity
26.5.4.1 Rank and Relevance
26.5.4.2 IR Diversity in Recommendation
26.5.4.3 Personalized Diversity
26.6 Empirical Metric Comparison
26.7 Conclusion
References
27 Cross-Domain Recommender Systems
27.1 Introduction
27.2 Formulation of the Cross-Domain Recommendation Problem
27.2.1 Definition of Domain
27.2.2 Cross-Domain Recommendation Tasks
27.2.3 Cross-Domain Recommendation Goals
27.2.4 Cross-Domain Recommendation Scenarios
27.3 Categorization of Cross-Domain Recommendation Techniques
27.4 Knowledge Aggregation for Cross-Domain Recommendations
27.4.1 Merging Single-Domain User Preferences
27.4.2 Mediating Single-Domain User Modeling Data
27.4.3 Combining Single-Domain Recommendations
27.5 Knowledge Linkage and Transfer for Cross-Domain Recommendation
27.5.1 Linking Domains
27.5.2 Sharing Latent Features by Domains
27.5.3 Transferring Rating Patterns Between Domains
27.6 Evaluation of Cross-Domain Recommender Systems
27.6.1 Data Partitioning
27.6.2 Metrics
27.6.3 Sensitivity Analysis
27.7 Practical Considerations in Cross-Domain Recommendation
27.8 Open Research Issues
References
28 Robust Collaborative Recommendation
28.1 Introduction
28.2 Defining the Problem
28.2.1 An Example Attack
28.3 Characterising Attacks
28.3.1 Basic Attacks
28.3.1.1 Random Attack
28.3.1.2 Average Attack
28.3.2 Low-Knowledge Attacks
28.3.2.1 Bandwagon Attack
28.3.2.2 Segment Attack
28.3.3 Nuke Attack Models
28.3.3.1 Love/Hate Attack
28.3.3.2 Reverse Bandwagon Attack
28.3.4 Informed Attack Models
28.3.4.1 Popular Attack
28.3.4.2 Probe Attack Strategy
28.3.4.3 Power User Attack
28.3.5 Obfuscated Attacks
28.4 Measuring Robustness
28.4.1 Evaluation Metrics
28.4.2 Push Attacks
28.4.3 Nuke Attacks
28.4.4 Informed Attacks
28.4.5 Attack Impact
28.5 Attack Detection
28.5.1 Evaluation Metrics
28.5.1.1 Impact on Recommender and Attack Performance
28.5.2 Single Profile Detection
28.5.2.1 Unsupervised Detection
28.5.2.2 Supervised Detection
28.5.3 Group Profile Detection
28.5.3.1 Neighbourhood Filtering
28.5.3.2 Detecting Attacks Using Profile Clustering
28.5.3.3 Hybrid Attack Detection
28.5.4 Detection Findings
28.6 Beyond Memory-Based Algorithms
28.6.1 Model-Based Recommendation
28.6.2 Privacy-Preserving Algorithms
28.6.3 The Influence Limiter and Trust-Based Recommendation
28.7 Robust Algorithms
28.7.1 Robust Matrix Factorisation (RMF)
28.7.2 Other Robust Recommendation Algorithms
28.8 Practical Countermeasures to Recommender System Attack
28.9 Conclusion
References
Index
有关键情节透露
