PRICAI 2021: Trends in Artificial Intelligence: 18th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2021, Hanoi, Vietnam, ... Part II (Lecture Notes in Computer Science)

This three-volume set, LNAI 13031, LNAI 13032, and LNAI 13033 constitutes the thoroughly refereed proceedings of the 18th Pacific Rim Conference on Artificial Intelligence, PRICAI 2021, held in Hanoi, Vietnam, in November 2021. The 93 full papers and 28 short papers presented in these volumes were carefully reviewed and selected from 382 submissions. PRICAI covers a wide range of topics in the areas of social and economic importance for countries in the Pacific Rim: artificial intelligence, machine learning, natural language processing, knowledge representation and reasoning, planning and scheduling, computer vision, distributed artificial intelligence, search methodologies, etc. Part II includes two thematic blocks: Natural Language Processing, followed by Neural Networks and Deep Learning. Preface Organization Contents – Part II Natural Language Processing A Calibration Method for Sentiment Time Series by Deep Clustering 1 Introduction 2 Related Work 3 Methods 3.1 Sentence Embedding 3.2 Representative Sampling 3.3 Sentiment Score Calibration 4 Experiment 4.1 Dataset 4.2 Baselines 4.3 Experimental Settings 4.4 Evaluation Metrics 4.5 Analysis of the Parameter Cluster Number 4.6 Compare with Random Sampling 5 Conclusion References A Weak Supervision Approach with Adversarial Training for Named Entity Recognition 1 Introduction 2 Related Work 3 Approach 3.1 Labeling Functions 3.2 WSAT: Weak Supervision Approach with Adversarial Training 4 Experimental Results 4.1 Dataset 4.2 Baselines 4.3 Results and Discussion 5 Conclusion References An Attention-Based Approach to Accelerating Sequence Generative Adversarial Nets 1 Introduction 2 Related Work 3 Our Approach 3.1 Attention-Based Discriminator 3.2 Attention to Rewards 3.3 Training of G 4 Experiments 4.1 Training Settings 4.2 Baselines 4.3 Evaluation Metrics 4.4 Synthetic Data Experiments 4.5 Dialogue Generation: DailyDialog 4.6 Internal Comparison Experiments 5 Conclusion and Future Work References Autoregressive Pre-training Model-Assisted Low-Resource Neural Machine Translation 1 Introduction 2 Background 3 Method 3.1 Partial Factorization Sequence Acquisition 3.2 NMT Model Integrated with Autoregressive Based XLNet 3.3 Knowledge Distillation Method 4 Experiments 4.1 Results and Analysis 4.2 Ablation Experiments 4.3 Case Study 5 Conclusion References Combining Improvements for Exploiting Dependency Trees in Neural Semantic Parsing 1 Introduction 2 Related Work 3 Three Improvements 3.1 Parent-Scaled Self-attention (PASCAL) 3.2 Syntax-Aware Word Representations (SAWRs) 3.3 Constituent Attention (CA) 4 Combining Improvements 5 Experiments 5.1 Datasets 5.2 Evaluation Metrics 5.3 Implementation Details 5.4 Results 5.5 Visual Analysis 6 Conclusion References Deep Semantic Fusion Representation Based on Special Mechanism of Information Transmission for Joint Entity-Relation Extraction 1 Introduction 2 Related Work 3 Task Definition and Tagging Scheme 4 The Proposed Model 4.1 Representations of Token and Relation 4.2 Deep Semantics Fusion 4.3 Triple Extraction 4.4 Training 5 Experiments 5.1 Dataset and Experimental Settings 5.2 Baselines and Evaluation Metrics 5.3 Experimental Results 6 Analysis 6.1 Ablation Study 6.2 Parameter Analysis 6.3 Analysis on Different Sentence Types 7 Conclusion References Exploiting News Article Structure for Automatic Corpus Generation of Entailment Datasets 1 Introduction 2 Methodology 2.1 NLI Datasets from News Articles 2.2 NewsPH-NLI 2.3 ELECTRA Pretraining 2.4 Benchmarking 2.5 Degradation Tests 3 Results and Discussion 3.1 Finetuning Results 3.2 Degradation Tests 3.3 Heuristics for Choosing Techniques for Low-Data Domains 4 Conclusion References Fake News Detection Using Multiple-View Text Representation 1 Introduction 2 Background and Related Work 2.1 Fake News Detection 2.2 Text Representation 3 Proposed Method 3.1 Multiple-View Text Representation 3.2 Integrated Stacking Model 4 Evaluation 4.1 Metrics 4.2 The Compared Fake News Detection Methods 4.3 Dataset 4.4 Experimental Results 5 Conclusions and Future Work References Generating Pseudo Connectives with MLMs for Implicit Discourse Relation Recognition 1 Introduction 2 Related Work 3 Our Model 3.1 Model Overview 3.2 Generating Connective Embeddings 3.3 Supervising the Generation of Connective 3.4 Relation Classification 4 Experiments 4.1 Dataset and Settings 4.2 Comparison Results 4.3 Discussion 5 Conclusion References Graph Convolutional Network Exploring Label Relations for Multi-label Text Classification 1 Introduction 2 Related Work 3 Proposed Method 3.1 Graph Convolutional Network Recap 3.2 Hop-Residual Graph Convolutional Networks for Multi-Label Classification 3.3 Loss Function 4 Experiments 4.1 Datasets 4.2 Evaluation Metrics 4.3 Baseline Methods and Setting Details 4.4 Analysis and Discussion 5 Conclusion References Improving Long Content Question Generation with Multi-level Passage Encoding 1 Introduction 2 Related Work 3 Architecture Overview 3.1 Encoder 3.2 Decoder 3.3 Reinforcement Learning 4 Experimental Setup 4.1 Datasets 4.2 Implementation Details 4.3 Evaluation Metrics 4.4 Baseline Methods 5 Results and Analysis 6 Conclusions References Learning Vietnamese-English Code-Switching Speech Synthesis Model Under Limited Code-Switched Data Scenario 1 Introduction 2 Vietnamese-English Code-Switching Synthesis Systems 2.1 Grapheme-to-Syllable-Based Code-Switching Synthesis 2.2 Speaker-Embedding Based Code-Switching Synthesis 2.3 Speaker-Embedding and Language-Embedding Based Code-Switching Synthesis 3 Experiments 3.1 Dataset 3.2 Training Setup 4 Evaluation and Discussion 4.1 Speech Quality Test 4.2 English Pronunciation Quality Test 4.3 English Pronunciation Accuracy Test 4.4 Discussion 5 Conclusion and Future Works References Multi-task Text Normalization Approach for Speech Synthesis 1 Introduction 2 Related Work 3 The Proposed Approach 3.1 Pre-processing Method 3.2 Prosodic Phrasing 3.3 Neural Machine Translation Model 4 Experiment 4.1 Dataset 4.2 Experimental Settings 5 Conclusion and Future Works References Performance-Driven Reinforcement Learning Approach for Abstractive Text Summarization 1 Introduction 2 Background 2.1 Related Works 2.2 Materials 3 PEARL: Performance-Driven Reinforcement Learning 3.1 Reward Function design and the problem of averaged score 3.2 FRouge: Rouge-Based REINFORCE Algorithm 3.3 DThreshold: Dynamic Cohesion Threshold for Document Unit 4 Experiments and Results 4.1 Implementation Detail 4.2 Result 5 Conclusion References Punctuation Prediction in Vietnamese ASRs Using Transformer-Based Models 1 Introduction 2 Related Work 3 A Proposed Model to Predict Punctuation 3.1 Problem Definition 3.2 A Proposed Architecture Using Pre-trained LMs with Boundary Context Information 4 Experiments 4.1 Datasets 4.2 Experimental Setups 4.3 Evaluation Metrics 4.4 Experimental Results 5 Conclusion References Rumor Detection on Microblogs Using Dual-Grained Feature via Graph Neural Networks 1 Introduction 2 Related Work 3 Du-FAGNN Rumor Detection Model 3.1 Problem Statement 3.2 Text-Level Feature Generation Module 3.3 Graph Neural Network Module 3.4 Pooling Module 4 Experiments and Analysis 4.1 Dataset 4.2 Experiment Settings 4.3 Baselines 4.4 Result Analysis 5 Conclusions References Short Text Clustering Using Joint Optimization of Feature Representations and Cluster Assignments 1 Introduction 2 Related Work 2.1 Short Text Clustering 2.2 Deep Neural Networks 3 Short Text Clustering Using Joint Optimization of Feature Representations and Cluster Assignments 3.1 Convolutional Neural Networks (CNN) 3.2 Clustering Loss 3.3 CAE-Based Short Text Clustering 4 Experiments 4.1 Datasets 4.2 Pre-trained Word Vectors 4.3 Experiment Setting 5 Results and Analysis 6 Conclusion References Soft-BAC: Soft Bidirectional Alignment Cost for End-to-End Automatic Speech Recognition 1 Introduction 2 Connectionist Temporal Classification 3 Proposed Approach 3.1 Soft Bidirectional Alignment Cost 3.2 Hybrid Soft-BAC Attention Architecture 4 Experimental Results 4.1 Datasets 4.2 Setup 4.3 Results 5 Conclusions References Span Labeling Approach for Vietnamese and Chinese Word Segmentation 1 Introduction 2 The Proposed Framework 2.1 Word Segmentation as Span Labeling Task for Vietnamese and Chinese 2.2 Post-processing Algorithm for Predicted Spans 2.3 Span Scoring Module 2.4 Encoder and Input Representation for VWS 2.5 Encoder and Input Representation for CWS 3 Experimental Settings 3.1 Datasets 3.2 Model Implementation 4 Results and Analysis 4.1 Main Results 4.2 Analysis 5 Conclusion References VSEC: Transformer-Based Model for Vietnamese Spelling Correction 1 Introduction 2 Related Work 3 Methodology 3.1 Problem Statement 3.2 Model Overview 3.3 Preprocessing 3.4 Tokenization 3.5 Transformer Model 4 Experimental Results 4.1 Dataset 4.2 Evaluation Metric 4.3 Experimental Setting 4.4 Main Results 4.5 Effect of Hyperparameter 4.6 Discussion 5 Conclusions References What Emotion Is Hate? Incorporating Emotion Information into the Hate Speech Detection Task 1 Introduction 2 Background and Related Work 3 The Proposed Approach 4 Experiment Design 4.1 Description of Datasets 5 Results 5.1 Further Analysis on the Counter-Fitted Embeddings (HateEmoEmb and AllEmoEmb) 6 Conclusion References Enhanced Named Entity Recognition with Semantic Dependency 1 Introduction 2 Related Work 3 Model 3.1 BiLSTM-CRF 3.2 Sem-BiLSTM-GCN-CRF 4 Experiment 4.1 Datasets 4.2 Experimental Setup 4.3 Main Results 4.4 Effect of Dependency Quality 5 Analysis 6 Conclusion References Improving Sentence-Level Relation Classification via Machine Reading Comprehension and Reinforcement Learning 1 Introduction 2 Overview 2.1 Problem Definition 2.2 Framework 3 Method 3.1 Data Processing 3.2 MRC Estimator 3.3 Instance Sampler 3.4 Reward Calculation Module 3.5 MRC Estimator Training 4 Experiment 4.1 Dataset 4.2 Parameter Settings 4.3 Baselines 4.4 Performance Comparison and Analysis 5 Conclusion References Multi-modal and Multi-perspective Machine Translation by Collecting Diverse Alignments 1 Introduction 2 Related Work 3 Methodology 3.1 The Framework of Our M3-CoDA 3.2 Image Feature Extraction Module 3.3 Diverse Alignments Module 3.4 Multi-perspective Ensemble Module 3.5 Loss Function 4 Experiment 4.1 Dataset and Evaluation Measure 4.2 Experimental Setup and Parameter Setting 4.3 Baselines 4.4 Experimental Results 5 Conclusion and Future Work References Simplifying Paragraph-Level Question Generation via Transformer Language Models 1 Introduction 2 Methodology 2.1 Data Preparation 2.2 Experiments 3 Results and Discussion 3.1 Evaluating Context-Copying 3.2 Failed Generations 3.3 Optimal Context Length 3.4 Answer-Awareness 4 Related Literature 5 Conclusion References Neural Networks and Deep Learning ABAE: Utilize Attention to Boost Graph Auto-Encoder 1 Introduction 2 Related Work 2.1 Graph Auto-Encoders 2.2 Subgraph Convolutional Networks 3 Proposed Improved AttSCNs 3.1 Insight of Subgraph Convolution 3.2 Attention-Based SCNs 4 Attention-Based Auto-Encoder 4.1 Inversed AttSCNs 4.2 Proposed ABAE 4.3 Loss of ABAE 5 Experiments 5.1 Dataset Description 5.2 Node Classification with AttSCNs 5.3 Link Prediction with AttSCN-Based Auto-Encoder 6 Conclusion and Future Work References Adversarial Examples Defense via Combining Data Transformations and RBF Layers 1 Introduction 2 Related Works 2.1 Attack Methods 2.2 Defense Methods 3 Proposed Method 3.1 Network Design 3.2 Loss Function 4 Experiments 4.1 Experimental Setting 4.2 Ablation Study 4.3 Defense Results 5 Conclusion References An Improved Deep Model for Knowledge Tracing and Question-Difficulty Discovery 1 Introduction 2 Related Work 2.1 Deep Knowledge Tracing 2.2 Self-Paced Learning 3 The Proposed Model 3.1 Problem Definition 3.2 SPDKT Model 3.3 Theoretical Analysis 4 Experiment 4.1 Data Sets 4.2 Comparison Methods 4.3 Student Performance Prediction 4.4 Analysis Knowledge State of Student 4.5 Analysis Difficult of Skills 5 Discussion and Conclusion References ARNet: Accurate and Real-Time Network for Crowd Counting 1 Introduction 2 Related Work 2.1 Crowd Counting and Density Estimation 2.2 Lightweight Networks 3 Proposed Method 3.1 Network Architecture 3.2 Loss Functions 4 Implementation Details 4.1 Ground Truth Generation 4.2 Data Augmentation 4.3 Training Process and Inference Process 4.4 Counting Performance Evaluation Metrics 5 Experiments 5.1 Datasets 5.2 Results and Analysis 6 Conclusion References Deep Recommendation Model Based on BiLSTM and BERT 1 Introduction 2 DSAM Model 2.1 Problem and Symbol Definition 2.2 DSAM Model Structure 3 Experiment 3.1 Dataset 3.2 Comparative Experiment 3.3 Experimental Detailed Settings 3.4 Experimental Analysis 4 Conclusion References GCMNet: Gated Cascade Multi-scale Network for Crowd Counting 1 Introduction 2 Related Works 2.1 Multi-scale Feature Extraction Methods 2.2 Multi-level Feature Fusion Methods 2.3 Feature-Wise Gated Convolution Methods 3 Proposed Algorithm 3.1 Overview of Network Architecture 3.2 Multi-scale Contextual Information Enhancement Module 3.3 Hopping Cascade 3.4 Gated Information Selection Delivery Module 4 Experiments 4.1 Datasets 4.2 Settings 4.3 Comparisons with the State-of-the-Art 4.4 Comparison of Density Map Quality 4.5 Ablation Study 5 Conclusion References GIAD: Generative Inpainting-Based Anomaly Detection via Self-Supervised Learning for Human Monitoring 1 Introduction 2 Related Work 2.1 Generative Reconstruction Approaches 2.2 Anomalous Combinations Detection 3 Methodology 3.1 Problem Formulation 3.2 Local Salient Region and Global Similarity Loss 3.3 Attention-Based Gaussian Weighting Anomaly Score 4 Experiments 4.1 Datasets 4.2 Experimental Setup 4.3 Results and Analysis 5 Conclusion References Heterogeneous Graph Attention Network for User Geolocation 1 Introduction 2 Related Work 2.1 User Geolocation 2.2 Heterogeneous Graph Neural Network 3 Preliminary 4 Approach 4.1 Heterogeneous Graph Construction 4.2 Heterogeneous Graph Attention Network 4.3 Context Attention Network 4.4 Geolocation Prediction 5 Experiment 5.1 Dataset 5.2 Baseline 5.3 Metrics 5.4 Implementation 5.5 Result 6 Conclusion References Hyperbolic Tangent Polynomial Parity Cyclic Learning Rate for Deep Neural Network 1 Introduction 2 Related Work 3 Hyperbolic Tangent Polynomial Parity Cyclic Learning Rate (HTPPC) 3.1 The HTPPC 3.2 LR Parity Range Determination 3.3 Cycle Period Stepsize 3.4 Curvature Parameter and Polynomial Shape Parameter 4 Experiments 4.1 Dataset 4.2 Experiment on CIFAR-10 and CIFAR-100 4.3 Experiment on Pascal VOC 4.4 Conclusion References Infrared Image Super-Resolution via Heterogeneous Convolutional WGAN 1 Introduction 2 Related Works 2.1 Generative Adversarial Networks 2.2 HetConv: Heterogeneous Kernel-Based Convolutions 3 HetSRWGAN 3.1 HetSRWGAN Architecture 3.2 Heterogeneous Kernel-Based Residual Block 3.3 Gradient Cosine Similarity Loss Function 4 Experiments and Evaluations 4.1 Training Details 4.2 Performance of the Final Networks 5 Conclusions References Knowledge Compensation Network with Divisible Feature Learning for Unsupervised Domain Adaptive Person Re-identification 1 Introduction 2 Related Work 2.1 Feature Distribution Alignment Methods 2.2 Image-Style Transformation Methods 2.3 Clustering with Fine-Tuning Methods 3 Proposed Method 3.1 Cross-Camera Data Augmentation (CDA) 3.2 Supervised Learning in the Source Domain 3.3 Knowledge Compensation Network in the Target Domain 3.4 Divisible Feature Learning with Outliers-Aware Clustering 3.5 Compensation-Guided Softened Loss for Domain Adaptation 3.6 Overall Loss and Algorithm 4 Experiments and Analysis 4.1 Implementation Details 4.2 Datasets and Evaluation Metrics 4.3 Ablation Experiments 4.4 Comparison with the State-of-the-Art Methods 5 Conclusion References LoCo-VAE: Modeling Short-Term Preference as Joint Effect of Long-Term Preference and Context-Aware Impact in Recommendation 1 Introduction 2 Related Work 2.1 General Recommendation Systems 2.2 Recommendation Systems Based on User Dynamic Preference 2.3 Context-Aware Recommendation Systems 3 Methodology 3.1 Notations and Problem Formulation 3.2 Model of LoCo-VAE 3.3 Objective Function 4 Experiments 4.1 Datasets 4.2 Baselines 4.3 Metrics 4.4 Parameter Settings 5 Results and Analysis 5.1 Exploratory Analysis 5.2 Performance Comparison 5.3 Case Study 6 Conclusion References Multi-scale Edge-Based U-Shape Network for Salient Object Detection 1 Introduction 2 Related Work 3 Methodology 3.1 Overview 3.2 UEN Module 3.3 Additional Down-Sampling Module 3.4 Loss Function 4 Experiment 4.1 Experimental Setting 4.2 Ablation Study 4.3 Comparison with Other State-of-the-Art Methods 5 Conclusion References Reconstruct Anomaly to Normal: Adversarially Learned and Latent Vector-Constrained Autoencoder for Time-Series Anomaly Detection 1 Introduction 2 Related Work 2.1 Traditional Anomaly Detection Methods 2.2 Deep-Learning Anomaly Detection Methods 3 Proposed Method: RAN 3.1 Problem Description 3.2 Imitate Anomaly Subsequences 3.3 Reconstruct Anomalies to Normal 4 Experiments 4.1 Experiments Setup 4.2 Comparison with State-of-the-Art Algorithms 4.3 Analysis of Performance 4.4 Ablation Study 5 Conclusion References Robust Ensembling Network for Unsupervised Domain Adaptation 1 Introduction 2 Related Work 3 Methodology 3.1 Overview 3.2 Robust Ensembling Network 3.3 Dual-Network Conditional Adversarial Learning 3.4 Consistency Constraint 3.5 Total Loss Function 4 Experiment 4.1 Experimental Setting 4.2 Results 4.3 Ablation Study and Visualization 4.4 Ablation Study and Visualization 5 Conclusion References SPAN: Subgraph Prediction Attention Network for Dynamic Graphs 1 Introduction 2 Related Work 3 Proposed Method 3.1 Bayesian Subgraph Sampling 3.2 Subgraph Prediction 4 Experiments 4.1 Subgraph Prediction 4.2 Subgraph Pattern Prediction 4.3 Model Analysis 4.4 Hyperparameter Analysis 5 Conclusion References WINVC: One-Shot Voice Conversion with Weight Adaptive Instance Normalization 1 Introduction 2 StarGAN-VC/VC2 2.1 Training Objectives 2.2 Generator Architectures 3 The Proposed Model 3.1 Workflow 3.2 The Generator with Weight Adaptive Instance Norm 3.3 The Speaker Encoder and the Discriminator 3.4 Training Objectives 4 Experiments 4.1 Datasets 4.2 Training Details 4.3 Subjective Evaluations 4.4 Objective Evaluations 5 Conclusions References Fusion Graph Convolutional Collaborative Filtering 1 Introduction 2 Related Works 2.1 Network Embedding 2.2 Graph-Based Collaborative Filtering 3 The Proposed Method 3.1 Overview 3.2 Graph Encoder 3.3 Decomposer 3.4 Predictor 4 Experiments 4.1 Experimental Settings 4.2 Performance Comparison(Q1) 4.3 Effect of DeepWalk(Q2) 4.4 Effects of ESIM Local Inference 5 Conclusions and Future Works References Multi-label Learning by Exploiting Imbalanced Label Correlations 1 Introduction 2 Approach 2.1 Image Representation Learning 2.2 Graph Convolutional Network 2.3 Multi-label Distribution Aware Margin Loss 3 Experiments 3.1 Algorithms for Comparison and Evaluation Metrics 3.2 Implementation Details 3.3 Comparisons with State-of-the Art Methods 4 Conclusion References Random Sparsity Defense Against Adversarial Attack 1 Introduction 2 Adversarial Threat 3 Proposed Method 3.1 Random Sparsity 3.2 Whitening 3.3 Combination of Random Sparsity and Whitening 4 Experiments 4.1 Steup 4.2 Comparison 5 Conclusion References Author Index