This book constitutes the refereed proceedings of the 20th Australasian Conference on Data Mining, AusDM 2022, held in Western Sydney, Australia, during December 12–15, 2022.
The 17 full papers included in this book were carefully reviewed and selected from 44 submissions. They were organized in topical sections as research track and application track.
Author(s): Laurence A. F. Park (editor), Heitor Murilo Gomes (editor), Maryam Doborjeh (editor), Yee Ling Boo (editor), Yun Sing Koh (editor), Yanchang Zhao (editor), Graham Williams (editor), Simeon Simoff (editor)
Publisher: Springer
Year: 2022
Language: English
Pages: 260
Preface
Organization
Contents
Research Track
Measuring Content Preservation in Textual Style Transfer
1 Introduction
2 Background and Motivation
2.1 Cosine Similarity
2.2 Disentanglement of Style and Content
2.3 The Style Invariant Embedding Assumption
3 Experiment
3.1 Dataset
3.2 Procedure
4 Results and Discussion
5 Conclusion
References
A Temperature-Modified Dynamic Embedded Topic Model
1 Introduction
2 Related Work
3 Methodology
3.1 The Dynamic Embedded Topic Model
3.2 The Proposed Approach: DETM-tau
4 Experiments and Results
4.1 Experimental Set-Up
4.2 Results
5 Conclusion
References
Measuring Difficulty of Learning Using Ensemble Methods
1 Introduction
2 Related Work
3 Instance Difficulty
3.1 Difficulty Measures
4 Experiments
5 Conclusion
References
Graph Embeddings for Non-IID Data Feature Representation Learning
1 Introduction
2 Background and Related Work
2.1 Classification Models and IID Assumption
2.2 Graph and Knowledge Graph Embeddings
2.3 Summary
3 Methodology
4 Dataset and Experiment Design
4.1 Dataset
4.2 Experiment Design
5 Results and Discussions
5.1 Imbalanced Data
5.2 Advantage of Using the Node2vec Embeddings
5.3 Evaluation and Discussion
6 Conclusions and Future Work
6.1 Traffic
6.2 Learn Feature Representation for Non-IID Data via Graph Embeddings
6.3 Future Work
References
Enhancing Understandability of Omics Data with SHAP, Embedding Projections and Interactive Visualisations
1 Introduction
2 Framework for Using SHAP to Optimise UMAP and PCA Input Data
2.1 Initial Visualisations from the UMAP and PCA Projection Methods
2.2 Explainable Machine Learning SHAP for Important Feature Selection
2.3 Final Optimised Visualisations
3 UMAP and PCA Visualisations
3.1 Datasets
3.2 How Do PCA, UMAP, and SHAP Work?
3.3 Similar Projection and Clustering Patterns Between PCA and UMAP
4 Rank and Select the Most Important Features with SHAP
5 Validation of the SHAP Results
6 Conclusion and Future Work
References
WinDrift: Early Detection of Concept Drift Using Corresponding and Hierarchical Time Windows
1 Introduction
2 Preliminaries
3 The WinDrift (WD) Method
3.1 Key Components
3.2 Step-by-Step Description
4 Experimental Results
4.1 Experimental Setup
4.2 Datasets
4.3 Numerical Results
5 Conclusion and Future Work
References
Investigation of Explainability Techniques for Multimodal Transformers
1 Introduction
2 Problem Definition
2.1 Quantifying Syntactic Grounding Through Label Attribution
2.2 Investigating Semantic Relationships Through Optimal Transport
3 Explainability Techniques
3.1 Label Attribution
3.2 Optimal Transport
4 A Case Study in VisualBERT Explainability
5 Conclusion
References
Effective Imbalance Learning Utilizing Informative Data
1 Introduction
2 Related Work
2.1 Sampling Method
2.2 Cost-Sensitive Methods
2.3 Ensemble Methods
2.4 Data Representation
3 Proposed Framework and Approach
3.1 Informative Samples Located
3.2 Extracting Information
3.3 Model Test
4 Experiments and Results
4.1 Results on General Test
4.2 Results on Different Reference Data
5 Conclusion
References
Interpretable Decisions Trees via Human-in-the-Loop-Learning
1 Introduction
2 Learning Classifiers Involving Dataset Visualisations
3 Experts Iteratively Construct Decision Trees
3.1 Using Parallel Coordinates
3.2 The Splits the User Shall Apply
3.3 Information that Supports Interaction
3.4 Visualising the Tree
3.5 Visualising Rules
4 Design of the Usability Evaluation
5 Results
5.1 Validity Threats
6 Conclusion
References
Application Track
A Comparative Look at the Resilience of Discriminative and Generative Classifiers to Missing Data in Longitudinal Datasets
1 Introduction
2 Background and Related Work
3 LoGAN: A GAN Based Longitudinal Classifier for Missing Data
3.1 The LoGAN Approach
4 Experiments
4.1 Dataset
4.2 Baseline Models
4.3 Experimental Setup and Model Design
4.4 Evaluation Criteria
5 Results and Discussion
5.1 Training Performance by Model Setting
5.2 Performance on Balanced Data
5.3 Performance on Imbalanced Data
6 Final Remarks
7 Conclusions
References
Hierarchical Topic Model Inference by Community Discovery on Word Co-occurrence Networks
1 Introduction
2 Related Work
3 Community Topic
3.1 Co-occurrence Network Construction
3.2 Community Mining
3.3 Topic Filtering and Term Ordering
3.4 Topic Hierarchy
4 Empirical Evaluation
4.1 Datasets
4.2 Preprocessing
4.3 Evaluation Metrics
5 Results
6 Conclusion
References
UMLS-Based Question-Answering Approach for Automatic Initial Frailty Assessment
1 Introduction
2 Related Work
3 The Proposed Approach
3.1 Discovery of UMLS Based Concepts
3.2 UMLS-Based Concept Selection Algorithm
3.3 Answering TFI Questionnaire Using UMLS-Based Concepts
3.4 Frailty Assessment
4 Experiment and Results
4.1 Dataset
4.2 Experiment Settings
4.3 Results
5 Discussion
6 Conclusion
References
Natural Language Query for Technical Knowledge Graph Navigation
1 Introduction
2 Related Work
3 Approach
4 Application
4.1 Overview of Maintenance KG
4.2 Neural Named Entity Recognition Ensemble
5 Results and Discussion
5.1 Ensemble NER Performance Analysis
5.2 Question Types Discussion
6 Conclusions and Future Work
References
Decomposition of Service Level Encoding for Anomaly Detection
1 Introduction
2 Algorithm and Notations
2.1 Input/Output Spaces
2.2 Algorithm
2.3 Defining Interval Sub Extreme (SE)
3 Results
3.1 Physiotherapy Service Levels
3.2 General Practitioner Service Levels
3.3 Psychiatric Service Levels
3.4 Discipline Comparison
4 Summary and Further Work
References
Improving Ads-Profitability Using Traffic-Fingerprints
1 Introduction
2 Algorithm
2.1 Step 1 – Clustering of Domains
2.2 Step 2 – Creating Blocking Rules
2.3 Step 3 – Reassigning Domains to Clusters
3 Offline Experiments
4 Online Experiments
5 Conclusions
References
Attractiveness Analysis for Health Claims on Food Packages
1 Introduction
2 Related Work
3 Consumer Preference Prediction of Health Claims
3.1 Dataset Collection
3.2 Prediction Model
4 Evaluation and Results
5 Case Studies
5.1 Specialised Terminology Factors
5.2 Sentiment and Metaphoricity Factors
6 The Deployment of the Proposed Attractiveness Analysis Model
7 Conclusion
References
SchemaDB: A Dataset for Structures in Relational Data
1 Introduction
1.1 Existing Datasets
1.2 Challenges of Flat Data
2 Dataset Curation
2.1 Collection and Filtration
2.2 Graph Transform and Canonisation
2.3 Heuristic Augmentation
3 Analytics
3.1 Summary Statistics
4 Research Potential and Applications
5 Conclusion
References
Author Index
