Machine Learning and Knowledge Discovery in Databases. Applied Data Science and Demo Track: European Conference, ECML PKDD 2020, Ghent, Belgium, ... Part V (Lecture Notes in Computer Science)

The 5-volume proceedings, LNAI 12457 until 12461 constitutes the refereed proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2020, which was held during September 14-18, 2020. The conference was planned to take place in Ghent, Belgium, but had to change to an online format due to the COVID-19 pandemic.The 232 full papers and 10 demo papers presented in this volume were carefully reviewed and selected for inclusion in the proceedings. The volumes are organized in topical sections as follows: Part I: Pattern Mining; clustering; privacy and fairness; (social) network analysis and computational social science; dimensionality reduction and autoencoders; domain adaptation; sketching, sampling, and binary projections; graphical models and causality; (spatio-) temporal data and recurrent neural networks; collaborative filtering and matrix completion. Part II: deep learning optimization and theory;active learning; adversarial learning; federated learning; Kernel methods and online learning; partial label learning; reinforcement learning; transfer and multi-task learning; Bayesian optimization and few-shot learning. Part III: Combinatorial optimization; large-scale optimization and differential privacy; boosting and ensemble methods; Bayesian methods; architecture of neural networks; graph neural networks; Gaussian processes; computer vision and image processing; natural language processing; bioinformatics. Part IV: applied data science: recommendation; applied data science: anomaly detection; applied data science: Web mining; applied data science: transportation; applied data science: activity recognition; applied data science: hardware and manufacturing; applied data science: spatiotemporal data. Part V: applied data science: social good; applied data science: healthcare; applied data science: e-commerce and finance; applied data science: computational social science; applied data science: sports; demo track. Preface Organization Contents – Part V Applied Data Science: Social Good Confound Removal and Normalization in Practice: A Neuroimaging Based Sex Prediction Case Study 1 Introduction 2 Sex Classification and Brain Size 3 Experimental Setup 3.1 Study Design 3.2 Confound Regression 3.3 Predictive Modelling 4 Data Samples and Features 4.1 Data Samples 4.2 Pre-processing and Feature Extraction 5 Results 5.1 Generalization Performance Estimates 5.2 Predictive Features 5.3 Out-of-Sample Performance 6 Conclusion References Energy Consumption Forecasting Using a Stacked Nonparametric Bayesian Approach 1 Introduction 1.1 Energy Forecasting Across Major Australian States 1.2 Challenges and Related Work 1.3 Contribution of Our Approach 2 Overview of Gaussian Process 3 Proposed Data Analytic Model for Energy Prediction 3.1 Feature Selection 3.2 Modelling for Energy Forecasting 4 Experimental Setup 4.1 Data Description and Preparation 4.2 Baselines and Other Machine Learning Models Used for Comparison 5 Results and Discussion 6 Conclusion References Reconstructing the Past: Applying Deep Learning to Reconstruct Pottery from Thousands Shards 1 Introduction 2 Related Work 3 Our Approach 3.1 Dataset Generation Method 3.2 Our Proposed Model 4 Experimental Setup 5 Result 6 Discussions and Limitations 7 Conclusion References CrimeForecaster: Crime Prediction by Exploiting the Geographical Neighborhoods' Spatiotemporal Dependencies 1 Introduction 2 Datasets 3 Problem Definition 4 Methodology 4.1 CrimeForecaster Framework Overview 5 Experiment 5.1 CrimeForecaster Experiment Data and Setup 5.2 Performance Comparison 5.3 Parameter Study 6 Related Work 7 Conclusion References PS3: Partition-Based Skew-Specialized Sampling for Batch Mode Active Learning in Imbalanced Text Data 1 Introduction 2 Related Work 3 PS3: Partition-Based Skew-Specialized Sampling for Batch-Mode Active Learning 3.1 Batch-Mode Imbalance Learning 3.2 Human in the Loop – Assessment of Labeling Effort for Hate-Speech Detection 4 Experimental Methodology 4.1 Datasets 4.2 Baseline Methods 4.3 Evaluation Criteria 5 Results and Discussion 5.1 Performance Evaluation 5.2 Computational Time Analysis 6 Conclusion and Future Direction References An Uncertainty-Based Human-in-the-Loop System for Industrial Tool Wear Analysis 1 Introduction 2 Foundations and Related Work 3 Methodology 3.1 Image Segmentation 3.2 Model: Dropout U-Net 3.3 Loss Function and Performance Evaluation Metric 3.4 Uncertainty Estimation 4 Experiments 4.1 Datasets, Preprocessing and Training Procedure 4.2 Performance Results 4.3 Evaluation: Uncertainty-Based Human-in-the-Loop System 5 Discussion and Outlook References Filling Gaps in Micro-meteorological Data 1 Introduction 2 Related Work 3 Filling Gaps 3.1 Architecture 3.2 Feed Forward Layer and Copy Task 3.3 Positional Encoding 3.4 Training 4 Experiments 4.1 Datasets 4.2 Evaluation 4.3 Neural Networks and Training 5 Results 5.1 Toy Case 5.2 Evapotranspiration Data 6 Conclusion References Lagrangian Duality for Constrained Deep Learning 1 Introduction 2 Preliminaries: Lagrangian Duality 3 Learning Constrained Optimization Problems 3.1 Motivating Applications 3.2 The Learning Task 3.3 Lagrangian Dual Framework for Constrained Optimization Problems 4 Learning Constrained Predictors 4.1 Motivating Applications 4.2 The Learning Task 4.3 Lagrangian Dual Framework for Constrained Predictors 5 Experiments 5.1 Constrained Optimization Problems 5.2 Constrained Predictor Problems 6 Related Work 7 Conclusions References Applied Data Science: Healthcare Few-Shot Microscopy Image Cell Segmentation 1 Introduction 2 Related Work 2.1 Cell Segmentation 2.2 Few-Shot Learning 3 Few-Shot Cell Segmentation 3.1 Problem Definition 3.2 Few-Shot Meta-learning Approach 3.3 Meta-learning Algorithm 3.4 Task Objective Functions 3.5 Fine-Tuning 4 Experiments 4.1 Implementation Details 4.2 Microscopy Image Databases 4.3 Assessment Protocol 4.4 Results Discussion 5 Conclusion References Deep Reinforcement Learning for Large-Scale Epidemic Control 1 Introduction 2 Related Work 3 Epidemiological Model 3.1 Intra-patch Model 3.2 Inter-patch Model 3.3 Calibration and Validation 4 Learning Environment 5 PPO Versus Ground Truth 6 Multi-district Reinforcement Learning 7 Discussion References GLUECK: Growth Pattern Learning for Unsupervised Extraction of Cancer Kinetics 1 Background 1.1 Tumor Growth and Its Implications 1.2 Mechanistic Models of Tumor Growth 1.3 Predictive Models of Tumor Growth 1.4 Peculiarities of Tumor Growth Data 2 Materials and Methods 2.1 Introducing GLUECK 2.2 Datasets 2.3 Procedures 3 Results 4 Conclusion References Automated Integration of Genomic Metadata with Sequence-to-Sequence Models 1 Introduction 2 Related Work 3 Approaches 3.1 Multi-label Classification Approach 3.2 Translation-Based Approach 4 Experiments 4.1 Datasets: GEO, Cistrome and ENCODE 4.2 Experimental Setup 4.3 Experiments 1 and 2 4.4 Experiment 3: Randomly Chosen GEO Instances 5 Conclusions and Future Work References Explaining End-to-End ECG Automated Diagnosis Using Contextual Features 1 Introduction 2 Related Works 3 Methodology 3.1 Case Study Method 3.2 Segmentation-Based Noise Insertion 4 Contextual Features for a Convolutional Network 4.1 Model Description 4.2 Model Evaluation 5 Discussion 6 Conclusions References Applied Data Science: E-Commerce and Finance A Deep Reinforcement Learning Framework for Optimal Trade Execution 1 Introduction 2 Limit Order Book and Market Microstructure 3 A DQN Formulation to Optimal Trade Execution 3.1 Preliminaries 3.2 Problem Formulation 3.3 DQN Architecture and Extensions 3.4 Experimental Methodology and Settings 4 Experimental Results 4.1 Data Sources 4.2 Training and Stability 4.3 Main Evaluation and Backtesting 5 Conclusion and Future Work A Hyperparameters References Detecting and Predicting Evidences of Insider Trading in the Brazilian Market 1 Introduction 2 Literature Review 3 Dataset Preparation 3.1 Impactful Events in 2017 3.2 Classifying Possible Evidences of Insider Trading 3.3 Expanding the Dataset for 2018 4 Recognising Suspicious Trades Before Events Unfold 4.1 Monitoring Negotiations 4.2 Predicting Relevant Events 5 Conclusion and Future Work References Mend the Learning Approach, Not the Data: Insights for Ranking E-Commerce Products 1 Introduction 2 Related Work 3 E-Com Dataset for LTR 3.1 Need of a New Dataset 3.2 Scope of the Dataset 3.3 Dataset Construction 4 Problem Formulation 5 Experiments 5.1 Experimental Setup 5.2 Comparison of CRM and Full-Info Approaches (RQ1) 5.3 Learning Progress with Increasing Number of Bandit Feedback (RQ2) 5.4 Effect of the DNN Architecture (RQ3) 6 Conclusion A Comparison of Counterfactual Risk Estimators B Choosing Hyperparameter References Multi-future Merchant Transaction Prediction 1 Introduction 2 Notations and Problem Definition 3 Model Architecture 3.1 Shape Sub-network 3.2 Scale Sub-network 4 Training Algorithm 5 Experiments 5.1 Description of the Datasets 5.2 Evaluation of Architecture Design Choice 5.3 Evaluation of the Multi-future Learning Scheme 6 Related Work 7 Conclusion References Think Out of the Package: Recommending Package Types for E-Commerce Shipments 1 Introduction 1.1 Contributions 2 Related Work 2.1 Existing Packaging Selection Process 2.2 Why Not Ordinal Regression? 2.3 Comparison with Standard Machine Learning Approaches for Package Planning 3 Two-Stage Approach for Optimal Package Selection 3.1 Stage 1: Estimating the Transit Damage Probability of a Product Given a Package Type 3.2 Stage 2: Identifying the Optimal Package Type for Each Product 4 Experimental Results 4.1 Calibration 4.2 Package Type Recommendation 4.3 Impact Analysis from Actual Shipment Data 5 Conclusion and Future Work References Topics in Financial Filings and Bankruptcy Prediction with Distributed Representations of Textual Data 1 Introduction 2 Literature Reviews 3 Data and Methods 3.1 Data 3.2 Latent Dirichlet Allocation 3.3 Topic Modelling in Embedding Spaces 3.4 Number of Topics Assessments 3.5 Bankruptcy Prediction Feature Sets 3.6 Experimental Setup 4 Experimental Results and Discussions 4.1 Number of Topics in MD&A 4.2 Topics in MD&A and Their Evolution 4.3 Predictive Performance 5 Conclusions References Why Did My Consumer Shop? Learning an Efficient Distance Metric for Retailer Transaction Data 1 Introduction 2 A Framework for Learning the Distance Metric 2.1 Weighing a Product Hierarchy 2.2 Formalization 2.3 Finding Optimal Weights 2.4 Finding Fixed Points 3 Experimental Evaluation 3.1 Experimental Settings 3.2 Weights Learning 3.3 Qualitative Cluster Analysis 3.4 Competitor Analysis 4 Related Work 5 Conclusion and Future Work References Fashion Outfit Generation for E-Commerce 1 Introduction 2 Related Work 3 Outfit Datasets 4 Methodology 4.1 Network Architecture 4.2 Outfit Scoring 4.3 Visual Feature Extraction 4.4 Title and Description Embeddings 4.5 Training 4.6 Outfit Generation Method 5 Evaluation 5.1 Train/test Split 5.2 Outfit Classification Results 5.3 Generated Outfit Evaluation 5.4 Style Space 6 Conclusion References Improved Identification of Imbalanced Multiple Annotation Intent Labels with a Hybrid BLSTM and CNN Model and Hybrid Loss Function 1 Introduction 2 Related Work 3 Methodology 3.1 Hybrid BLSTM-2DCNN Model 3.2 Hybrid Loss Function 4 Experimental Framework 4.1 Dataset 4.2 Experiment Setting 5 Results and Discussion 6 Conclusion References Measuring Immigrants Adoption of Natives Shopping Consumption with Machine Learning 1 Introduction 2 Related Work 3 Problem Formulation 4 Trends of Immigrants Native Consumption Adoption 4.1 Modeling Customer Shopping Behavior 4.2 Learning and Measuring Native Consumption Adoption (NCA) 4.3 Grouping and Monitoring NCA Trends of Foreign Customers 5 Experiments 5.1 Data Analysis 5.2 Machine Learning and Classification Performance Analysis 5.3 Trends of Native Consumption Adoption Analysis 6 Conclusion References Applied Data Science: Computational Social Science Model Bridging: Connection Between Simulation Model and Neural Network 1 Introduction 2 Related Works 2.1 Simulator Calibration 2.2 Application of Kernel Mean Embedding 2.3 Distribution Regression 3 Proposed Framework: Model Bridging 3.1 Problem Setting, Assumption, and Usage of Model Bridging 3.2 Distribution-to-Distribution Regression 3.3 Input of Distribution-to-Distribution Regression 4 Experiment 4.1 Common Setting of Experiments 4.2 Experiment with Simple Production Simulator 4.3 Experiment with Realistic Production Simulator 4.4 Experiment with Simulator for Fluid Dynamics 5 Discussion 6 Conclusion References Semi-supervised Multi-aspect Detection of Misinformation Using Hierarchical Joint Decomposition 1 Introduction 2 Problem Formulation 3 Proposed Methodology 3.1 Multi-aspect Article Representation 3.2 Hierarchical Decomposition 3.3 Semi-supervised Article Inference 4 Experimental Evaluation 4.1 Dataset Description 4.2 Baselines for Comparison 4.3 Comparing with Baselines 5 Related Work 6 Conclusion References A Deep Dive into Multilingual Hate Speech Classification 1 Introduction 2 Related Works 3 Dataset Description 4 Experiments 4.1 Embeddings 4.2 Models 4.3 Hyperparameter Optimization 5 Results 5.1 Monolingual Scenario 5.2 Multilingual Scenario 5.3 Possible Recipes Across Languages 6 Discussion and Error Analysis 6.1 Interpretability 6.2 Error Analysis 7 Conclusion References Spatial Community-Informed Evolving Graphs for Demand Prediction 1 Introduction 2 Problem Statement 3 The Proposed Model 3.1 Learning Phase 3.2 Prediction Phase 4 Experiments 4.1 Datasets 4.2 Experiment Settings 4.3 Prediction Results 4.4 Capability of Predicting New Stations 4.5 Visualization of Intra-weights and Inter-weights 4.6 Parameter Analysis 5 Related Work 5.1 Demand Prediction 5.2 Spatial-Temporal Computing 6 Conclusion References Applied Data Science: Sports SoccerMix: Representing Soccer Actions with Mixture Models 1 Introduction 2 Methodology 2.1 Describing Actions 2.2 Grouping Actions with Mixture Models 2.3 Distributions of Locations and Directions 2.4 Fitting a Mixture Model to the Data 2.5 Practical Challenges 2.6 Capturing Playing Style with SoccerMix 3 Experiments 3.1 De-anonymizing Players 3.2 Comparing the Playing Style of Players 3.3 Comparing the Playing Style of Teams 3.4 Capturing the Defensive Playing Style of Teams 3.5 Case Study: How Liverpool Lost the Title to Manchester City in a Single Game 4 Related Work 5 Conclusion References Automatic Pass Annotation from Soccer Video Streams Based on Object Detection and LSTM 1 Introduction 2 Related Work 3 Pass Detection Problem 4 PassNet 4.1 Feature Extraction 4.2 Object Detection 4.3 Sequence Classification 5 Data Sets 5.1 Pass Annotation Application 6 Experiments 6.1 Results 7 Conclusion References SoccerMap: A Deep Learning Architecture for Visually-Interpretable Analysis in Soccer 1 Introduction 2 Related Work 3 A Deep Model for Interpretable Analysis in Soccer 3.1 The Reasoning Behind the Choice of Layers 3.2 Learning from Single-Location Labels 3.3 Spatial and Contextual Channels from Tracking Data 4 Experiments and Results 4.1 Dataset 4.2 Benchmark Models 4.3 Experimental Framework 4.4 Results 5 Practical Applications 5.1 Adapting SoccerMap for the Estimation of Pass Selection Likelihood and Pass Value 5.2 Assessing Optimal Passing and Location 5.3 Team-Based Passing Selection Tendencies 6 Discussion and Future Work References Stop the Clock: Are Timeout Effects Real? 1 Introduction 2 Related Work 3 The Causality Framework 4 The Causal Effect of Timeout 4.1 Short-Term Momentum Change 4.2 The Causal Model 4.3 Data 4.4 Matching 5 Experimental Results 5.1 Matching Results 5.2 Timeout Effect 6 Conclusion References Demo Track Deep Reinforcement Learning (DRL) for Portfolio Allocation 1 Introduction 2 Method Used 3 Target Users and Future Extension References Automated Quality Assurancepg for Hand-Held Tools via Embedded Classification and AutoML 1 Introduction 2 System Architecture 3 About the Demonstration References Massively Distributed Clustering via Dirichlet Process Mixture 1 Introduction 2 Distributed Clustering via Dirichlet Process Mixture 3 High Dimensional Data Distributed Dirichlet Clustering 4 DC-DPM and HD4C in Spark 5 Demonstration References AutoRec: A Comprehensive Platform for Building Effective and Explainable Recommender Models 1 Introduction 2 Overview of the System 3 System Demonstration References multi-imbalance: Open Source Python Toolbox for Multi-class Imbalanced Classification 1 Introduction 2 Overview of the Package 3 Project Design 4 An Example of Use 5 Conclusions References VisualSynth: Democratizing Data Science in Spreadsheets 1 Introduction 2 Interacting Using Colors 3 Use Case: Ice Cream Sales Auto-completion 4 Architecture References FireAnt: Claim-Based Medical Misinformation Detection and Monitoring 1 Introduction 2 System Overview 2.1 Data Acquisition 2.2 Claim-Based Article Veracity Determination 2.3 Results Presentation and Explanation References GAMA: A General Automated Machine Learning Assistant 1 Introduction 2 System Overview 3 Related Work 4 Conclusion References Instructional Video Summarization Using Attentive Knowledge Grounding 1 Introduction 2 Problem Formulation 3 Proposed Model 4 Data Collection and Demonstration References PrePeP: A Light-Weight, Extensible Tool for Predicting Frequent Hitters 1 Introduction 2 Mining Discriminative Subgraphs 3 Predicting Frequent Hitters 4 Visualizing Graphs Supporting the Decision 5 Conclusion References Author Index