Inhalt

•  
   Titlepage
•  
   Abstract
•  
   Table of Contents
•  
   List of Figures
•  
   List of Tables
•  
   I Introduction
•  
   1 Introduction
•  
   1.1 Motivating Use Cases
•  
   1.2 Goal and Challenges
•  
   1.2.1 Clustering of Large Datasets
•  
   1.2.2 Clustering of Continuous Data Streams
•  
   1.2.3 Classifying of Concept Drifting Time Series Data
•  
   1.2.4 Noisy Data
•  
   1.3 Research Contributions of this Dissertation
•  
   1.4 Structure and Outline of this Thesis
•  
  2 Fundamentals of This Work
•  
   2.1 From the Categorization Idea to Event Clustering: Definition and Development
•  
   2.1.1 Categorization in Philosophy — The Classical View
•  
   2.1.2 Categorization in Cognitive Psychology — The Prototype View
•  
   2.1.3 Event Clustering Characterization
•  
   2.2 Characterization of an Event
•  
   2.2.1 Event Definition in Philosophy
•  
   2.2.2 Event Definition in Cognition and Psychology
•  
   2.2.3 Events in Recent Literature of Machine Learning and Information Retrieval
•  
   2.2.4 Discussion and Definition
•  
   3 Foundations and Related Work
•  
   3.1 Classification
•  
   3.2 Clustering
•  
   3.3 Distance Functions
•  
   3.4 Knowledge-based Clustering
•  
  3.5 Large-scale Processing and Scalability
•  
   3.5.1 Task-based Techniques
•  
   3.5.2 Data-based Techniques
•  
   3.5.3 Candidate Retrieval
•  
   3.5.4 Stream Data
•  
   3.6 New Event Detection
•  
   3.6.1 Statistical Approaches
•  
   3.6.2 Unsupervised Approaches
•  
   3.6.3 Supervised Approaches
•  
   3.7 Event Identification and Detection
•  
   4 Event Clustering Dataset
•  
   4.1 Creation and Collection of the Dataset
•  
   4.1.1 Fetching of Metadata
•  
   4.1.2 Fetching of Uploader Information
•  
   4.1.3 Fetching of Picture Files
•  
  4.2 Labeling of the Data — Creation of the Gold Standard
•  
   4.2.1 Usage of Social Event Calendars for Data Labeling
•  
   4.2.2 Fetching of Event Information from Upcoming and Last.fm
•  
   4.2.3 Labeling Process
•  
   4.3 Dataset Statistics
•  
   4.3.1 Data Quality
•  
   4.3.2 License Constraints
•  
   4.3.3 Data Point Distribution
•  
   4.3.4 Dataset Representation Format and Schema
•  
  4.4 Applications of the Dataset
•  
   4.4.1 MediaEval 2013
•  
   4.4.2 Further Applications
•  
   4.4.3 Evaluation Proposal for Comparability
•  
   II Supervised Single-Pass Clustering with the Event-based Stream Classification Framework
•  
   5 System Description of the Stream Classification Framework for a Single-Pass Setting
•  
   5.1 Problem Statement
•  
   5.2 Overview of the Clustering Framework
•  
  5.3 Candidate Retrieval Strategies
•  
   5.3.1 Measurements for Performance and Effectiveness
•  
   5.3.2 Candidate Retrieval Strategies
•  
  5.4 Pairwise Feature Extraction
•  
   5.4.1 Temporal Features
•  
   5.4.2 Geographical Features
•  
   5.4.3 Textual Features
•  
   5.4.4 Document-Event Similarity Vector
•  
   5.5 Scoring and Ranking — Learning Similarity Functions
•  
   5.5.1 Problem Formulation using a Support Vector Machine
•  
   5.5.2 Problem Formulation as a Decision Tree Classification Problem
•  
   5.6 New Event Detection
•  
  6 Experimental Setup and Results of the Supervised Single-Pass Classification
•  
   6.1 Definition of Evaluation Measures
•  
  6.2 Optimizing Candidate Retrieval
•  
   6.2.1 Experimental Settings
•  
   6.2.2 Results
•  
   6.2.3 Conclusion
•  
  6.3 Learning Similarity Functions
•  
   6.3.1 Experimental Settings
•  
   6.3.2 Results
•  
   6.3.3 Conclusion
•  
  6.4 New Event Detection
•  
   6.4.1 Experimental Settings
•  
   6.4.2 Results
•  
   6.4.3 Conclusion
•  
  6.5 Framework as a Whole — Results and Comparison
•  
   6.5.1 Training and Optimization of the System Parts
•  
   6.5.2 Baselines
•  
   6.5.3 Overall System Performance
•  
   6.6 Conclusions
•  
   III Multi-pass Stream Clustering
•  
   7 System Description of the Stream Classification Framework for a Multi-Pass Setting
•  
   7.1 Problem Statement
•  
   7.2 System Overview
•  
   7.3 Multi-pass Requirements and Challenges
•  
   7.3.1 Number of Passes
•  
   7.3.2 Influence on Framework Settings
•  
   7.4 Multi-pass Strategies
•  
  8 Experimental Setup and Results of Supervised Multi-Pass Clustering
•  
   8.1 Analysis of First-Pass Strategies
•  
  8.2 Gold Standard Preparation for the Second Pass
•  
   8.2.1 Quality Issues in the Preparation Process
•  
   8.2.2 Creation of the Gold Standard for the Second Pass
•  
  8.3 Optimization of the Classification Framework Steps for the Second Pass
•  
   8.3.1 Candidate Retrieval
•  
   8.3.2 Features for Similarity Function Learning and New Event Detection
•  
   8.4 Clustering Framework in Two-Pass Mode — Optimization
•  
   8.4.1 Exhaustive Search for Optimal Features in Scoring, Ranking, and New Event Detection
•  
   8.4.2 Results of the Exhaustive Search
•  
   8.4.3 Optimization of Candidate Retrieval Strategy
•  
   8.5 Results of the Clustering Framework used in Two-Pass Mode
•  
   8.6 Conclusions
•  
   IV Concluding Remarks
•  
  9 Remarks and Comparison of Clustering Approaches
•  
   9.1 Prerequisites for Event Clustering
•  
   9.2 Reflection on Multi-Pass Clustering in a Stream-based Setting
•  
   9.3 Comparison with Other Approaches
•  
   10 Conclusion
•  
   V Appendix
•  
   Glossary
•  
   Acronyms
•  
   Bibliography