PRICAI 2022: Trends in Artificial Intelligence: 19th Pacific Rim International Conference on Artificial Intelligence, PRICAI 2022, Shanghai, China, ... Part II (Lecture Notes in Computer Science)

This three-volume set, LNAI 13629, LNAI 13630, and LNAI 13631 constitutes the thoroughly refereed proceedings of the 19th Pacific Rim Conference on Artificial Intelligence, PRICAI 2022, held in Shangai, China, in November 10–13, 2022. The 91 full papers and 39 short papers presented in these volumes were carefully reviewed and selected from 432 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. Preface Organization Contents – Part II Knowledge Representation and Reasoning Moderately-Balanced Representation Learning for Treatment Effects with Orthogonality Information 1 Introduction 2 Preliminaries 3 Method 3.1 Orthogonality Information 3.2 The Proposed Framework 4 Experiments 4.1 Dataset Description 4.2 Performance Measurement and Experimental Settings 4.3 Results Analysis 4.4 Simulation Study 5 Related Work 6 Conclusion References Source-Free Implicit Semantic Augmentation for Domain Adaptation 1 Introduction 2 Related Work 2.1 Unsupervised Domain Adaptation (UDA) 2.2 Source-Free Domain Adaptation (SFDA) 3 Method 3.1 Overall Scheme 3.2 Source Class Prototype Generation 3.3 Target Class Prototype Generation Based on Pseudo-labelling Strategy 3.4 Source-Free Semantic Augmentation Adaptation Based on Generated Class Prototypes 3.5 Diversity and Discriminability Analysis During Source-Free Adaptation 3.6 Theoretical Insight 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Results 4.4 Robustness Analysis 4.5 Ablation Study 5 Conclusions References Role-Oriented Network Embedding Method Based on Local Structural Feature and Commonality 1 Introduction 2 Related Work 3 Methodology 3.1 Notions 3.2 Model 3.3 Complexity Analysis 4 Experiments 4.1 Datasets 4.2 Baseline 4.3 Experiment Settings 4.4 Visualization 4.5 Role-Oriented Node Classification 4.6 Top-k Similarity Search 5 Conclusion References Dynamic Refining Knowledge Distillation Based on Attention Mechanism 1 Introduction 2 Related Work 3 Methodology 3.1 Overall Structure of DRKD 3.2 KE Blocks 3.3 Loss Function 4 Experiments 4.1 Experiments on Benchmark Datasets 4.2 Comparison with Other Methods 4.3 Ablation Experiments 4.4 Refining Factor 5 Conclusion References Entity Representation by Neighboring Relations Topology for Inductive Relation Prediction 1 Introduction 2 Related Work 3 Methods 3.1 Relations Topology Module 3.2 Information Aggregation Module 3.3 Scoring Network and Loss Function 4 Experiments 4.1 Experimental Setup 4.2 Results and Analysis 4.3 Ablation Study 4.4 Performance with Different Number of Hops 5 Conclusion References Entity Similarity-Based Negative Sampling for Knowledge Graph Embedding 1 Introduction 2 Related Work 3 Entity Similarity-Based Negative Sampling Framework 3.1 Problem Definition 3.2 The Analysis of Quality Negative Samples 3.3 ESNS Negative Sampling Method 4 Experiments 4.1 Experimental Setup 4.2 Main Results 4.3 SPL Loss Function 5 Conclusion References Label Enhancement Using Inter-example Correlation Information 1 Introduction 2 Related Work 2.1 Label Propagation Based LE Algorithm (LP) 2.2 The LE Algorithm Based on Manifold Learning (ML) 2.3 Privileged Label Enhancement Method with Multi-label Learning (PLEML) 3 Methodology 3.1 Formulation of Label Enhancement 3.2 The LEEC Algorithm 3.3 Optimization 4 Experiments 4.1 DataSets 4.2 Evaluation Measures 4.3 Experimental Setting 4.4 Experimental Results 5 Conclusion References Link Prediction via Fused Attribute Features Activation with Graph Convolutional Network 1 Introduction 2 Related Work 3 AAGCN Model 3.1 Problem Formulation 3.2 Node Attribute Activation 3.3 Relation Features 3.4 Node Features 3.5 Features Integration 4 Experiments 4.1 Experimental Settings 4.2 Results and Discussion 4.3 Ablation Experiment 5 Conclusion References Multi-subspace Attention Graph Pooling 1 Introduction 2 Related Work 3 Methodology 3.1 Preliminaries 3.2 Pooling via Multi-subspace Attention 3.3 Deep Neural Networks with Pooling 4 Experiments and Analysis 4.1 Datasets and Baselines 4.2 Setup 4.3 Main Results 4.4 Ablation Study 4.5 More Analysis 5 Conclusion References Learning Temporal and Spatial Embedding for Temporal Knowledge Graph Reasoning 1 Introduction 2 Related Work 2.1 Static Knowledge Graph Embeddings 2.2 Temporal Knowledge Graph Embeddings 2.3 Embedding with Auxiliary Information 3 Spatial-Temporal Network 3.1 Notations 3.2 Model Components 3.3 Parameter Learning and Inference 4 Experiments 4.1 Experimental Setup 4.2 Results 4.3 Ablation Study 5 Conclusion References Natural Language Processing M2FNet: Multi-granularity Feature Fusion Network for Medical Visual Question Answering 1 Introduction 2 Related Work 2.1 Vision Question Answer 2.2 Meta Learning 3 Method 3.1 Overview 3.2 Image Feature Extraction 3.3 Multi-granularity Question Feature Extraction 3.4 Attention-Based Multi-granularity Fusion Module 3.5 Answer Prediction and Model Training 4 Experiments 4.1 Datasets and Metrics 4.2 Experimental Setup 4.3 Model Comparisons 4.4 Ablation Study 4.5 Qualitative Evaluation 5 Conclusion References Noise-Robust Semi-supervised Multi-modal Machine Translation 1 Introduction 2 Related Work 2.1 Multi-modal Machine Translation 2.2 Semi-supervised Machine Translation 2.3 Noise-Robust Alignment 3 Methodology 3.1 Noise-Robust Semantic Alignment 3.2 Semi-supervised Learning. 4 Experiment 4.1 Experimental Setup 4.2 Baselines 4.3 Overall Comparison (RQ1) 4.4 Ablation Study (RQ2) 4.5 Case Study (RQ3) 5 Conclusion and Future Work References SETFF: A Semantic Enhanced Table Filling Framework for Joint Entity and Relation Extraction 1 Introduction 2 Related Work 3 The Framework 3.1 Table Item Definition 3.2 Methodology 4 Experiments 4.1 Experimental Settings 4.2 Main Results 4.3 Ablation Study 4.4 Analysis on Hyper Parameter m 4.5 Analysis on Different Sentence Types 5 Conclusion References PEKIN: Prompt-Based External Knowledge Integration Network for Rumor Detection on Social Media 1 Introduction 2 Related Works 2.1 Rumor Detection 2.2 Prompt-Tuning 2.3 External Knowledge Integration 3 Method 3.1 Prompt Learning 3.2 External Knowledge Extraction 3.3 Rumor Classification 4 Experiments 4.1 Datasets 4.2 Experimental Setting and Data Preprocessing 4.3 Compared Methods 4.4 Primary Results 4.5 Prompt Strategy Analysis 4.6 Ablation Study 5 Conclusion and Future Works References Entity-Aware Social Media Reading Comprehension 1 Introduction 2 Related Work 3 Approach 3.1 Text Preprocessing 3.2 Entity-Aware Encoding Grounded on Multi-task Learning 3.3 Generating Answers 4 Experimentation 4.1 Data, Evaluation and Hyperparameter Settings 4.2 State-of-the-art SMRC Models for Comparison 4.3 Main Results 4.4 Ablation Study 4.5 Effects of Different Pretrained Models for Transfer 4.6 Utility of NER Toolkits 4.7 Error Analysis 5 Conclusion References Aspect-Based Sentiment Analysis via Virtual Node Augmented Graph Convolutional Networks 1 Introduction 2 Related Work 3 Proposed Model 3.1 Task Definition 3.2 Node Embeddings Initialization 3.3 Construction of the Virtual Node Augmented Graph 3.4 Virtual Node Augmented Graph Convolutional Network 3.5 Model Training 4 Experiments 4.1 Datasets and Experiment Settings 4.2 Comparison Baselines 4.3 Comparison Results 4.4 Ablation Analysis 4.5 Parameter Sensitivity 4.6 Case Study 5 Conclusion References Bidirectional Macro-level Discourse Parser Based on Oracle Selection 1 Introduction 2 Related Work 3 Bidirectional Discourse Parser on Oracle Selection 3.1 Basic Parser 3.2 Decision Maker 3.3 Oracle Selection 4 Experimentation 4.1 Dataset and Experimental Settings 4.2 Experimental Results 4.3 Analysis on Different Document Lengths 4.4 Results on English RST-DT 4.5 Error Analysis 4.6 Ablation Analysis 5 Conclusion References Evidence-Based Document-Level Event Factuality Identification 1 Introduction 2 Related Work 2.1 Event Factuality Identification 2.2 Evidence-Based Fact Checking 3 Corpus Annotation 3.1 Event 3.2 Document-Level Event Factuality 3.3 Evidential Sentences 3.4 Statistics 4 Methodology 4.1 Evidential Sentence Selection 4.2 Event Factuality Identification 5 Experimentation 5.1 Experimental Settings 5.2 Baselines 5.3 Overall Results 5.4 Ablation Study 5.5 Robustness Study 5.6 Error Analysis 6 Conclusion References Named Entity Recognition Model of Power Equipment Based on Multi-feature Fusion 1 Introduction 2 Related Work 2.1 Named Entity Recognition in Electric Power Domain 2.2 Data Augmentation 3 Methods 3.1 Data Augmentation Using Domain Dictionary 3.2 Fusion Embedding Layer 4 BiLSTM Layer 4.1 CRF Layer 5 Experiments and Results 5.1 Dataset 5.2 Evaluation Indicators 5.3 Experimental Design and Parameters 5.4 Experimental Results 6 Conclusions References Improving Abstractive Multi-document Summarization with Predicate-Argument Structure Extraction 1 Introduction 2 Preliminaries 3 Proposed Model 3.1 PAS Extraction 3.2 Summary Generation 4 Experiments 4.1 Datasets and Evaluation Metrics 4.2 Baselines 4.3 Experimental Setup 4.4 Results 5 Related Work 6 Conclusion References A Structure-Aware Method for Cross-domain Text Classification 1 Introduction 2 Related Works 2.1 Cross-domain Text Classification with Independence Assumption 2.2 Cross-domain Text Classification Methods with Graphs 3 Preliminary Knowledge 4 Our Method 4.1 Problem Definition 4.2 The Construction of Heterogeneous Graphs 4.3 Representations Learning for Invariant Structure 4.4 Classifier Training with Features and Structure 5 Experiment 5.1 Setup 5.2 Results and Analysis 5.3 Ablation Experiment 5.4 Parameter Discussion 6 Conclusions References SICM: A Supervised-Based Identification and Classification Model for Chinese Jargons Using Feature Adapter Enhanced BERT 1 Introduction 2 Related Work 3 Methodology 3.1 Character Feature Extraction Module 3.2 Feature Adapter Enhanced BERT Module 3.3 Global Attention Layer 4 Experiments 4.1 Dataset Construction 4.2 Sequence Labeling Baseline Model Comparison Experiment 4.3 Unsupervised Methods Comparison Experiment 5 Conclusion References HS2N: Heterogeneous Semantics-Syntax Fusion Network for Document-Level Event Factuality Identification 1 Introduction 2 Related Work 3 Methodology 3.1 Semantics-Syntax Fusion 3.2 Heterogeneous Graph 4 Experimentation 4.1 Datasets 4.2 Experimental Settings 4.3 Baselines 4.4 Result and Analysis 4.5 Ablation Study 5 Conclusion References Pay Attention to the ``Tails'': A Novel Aspect-Fusion Model for Long-Tailed Aspect Category Detection 1 Introduction 2 Related Work 2.1 Aspect Category Detection 2.2 Long-Tailed Distribution in Multi-label Classification 3 Methodology 3.1 Problem Formulation 3.2 Proposed Model 3.3 Loss Function 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Experimental Results and Analysis 4.4 Ablation Study 5 Conclusion References Choice-Driven Contextual Reasoning for Commonsense Question Answering 1 Introduction 2 Our Approach 2.1 Context Retrieval 2.2 Contextualized Encoding 2.3 Similarity-Based Differentiation 2.4 Confidence Aggregation 3 Experiments 3.1 Comparison on CommonsenseQA 3.2 Ablation Study 3.3 The Effectiveness of Choice-Driven Strategy 4 Related Work 5 Conclusions References Implicit Discourse Relation Recognition Based on Multi-granularity Context Fusion Mechanism 1 Introduction 2 Related Work 3 Method 3.1 Word Interaction Between Argument and Context (WIBAC) 3.2 Sentence Interaction Between Argument and Context (SIBAC) 3.3 Residual Network 4 Experiments 4.1 Dataset and Evaluation Metrics 4.2 Hyperparameter Setting 4.3 Results 4.4 Ablation Experiments 4.5 Effects and Analysis of Different Context Fusion Mechanisms 4.6 Selection of Critical Text Sequence Length 5 Conclusion References Chinese Medical Named Entity Recognition Using External Knowledge 1 Introduction 2 Related Work 3 Methods 3.1 Template-Based Method 3.2 Multi-feature Layer 3.3 Label Prediction 4 Experiments 4.1 Dataset 4.2 Setting 4.3 Evaluation 4.4 Results 4.5 Ablation Study 5 Conclusion References Neural Networks and Deep Learning Trajectory Prediction with Heterogeneous Graph Neural Network 1 Introduction 2 Related Work 3 Model Design 3.1 Problem Formulation 3.2 Feature Extraction Module 3.3 Scene Context Aware Destinations Prediction 3.4 Heterogeneous Graph Message Exchange 3.5 Trajectory Prediction 4 Experiments 4.1 Dataset 4.2 Experimental Settings 4.3 Baselines 4.4 Quantitative Evaluation 4.5 Ablation Study 4.6 Qualitative Evaluation 5 Conclusion References EEF1-NN: Efficient and EF1 Allocations Through Neural Networks 1 Introduction 2 Related Work 3 Preliminaries 4 Our Approach: EEF1-NN 4.1 Optimization Problem 4.2 EEF1-NN: Lagrangian Loss Function 4.3 Network Details 4.4 Training Details 5 Experiments and Results 5.1 Ablation Study 5.2 Experiment Details and Observations 6 Conclusion References Weighted Adaptive Perturbations Adversarial Training for Improving Robustness 1 Introduction 2 Related Work 2.1 Adversarial Attacks 2.2 Standard Adversarial Training 3 Weighted Adaptive Perturbation Adversarial Training 3.1 Motivations of WAPAT 3.2 Learning Objective of WAPAT 3.3 Realization of WAPAT 4 Experiments 4.1 Experimental Setup 4.2 Effectiveness of WAP-Attack 4.3 Evaluation Results for WAPAT 5 Conclusion and Future Work References Improved Network Pruning via Similarity-Based Regularization 1 Introduction 2 Related Work 3 Method 3.1 Notations 3.2 Focus Coefficient 3.3 Regularization 4 Improvements to Dense Networks 5 Improvements to Sparse Networks 5.1 Data Augmentation Hurts Network Pruning 5.2 Explicit Regularizers 5.3 Transformer Architecture 6 Conclusion References Dynamic-GTN: Learning an Node Efficient Embedding in Dynamic Graph with Transformer 1 Introduction 2 Related Works 3 Continuous-Time Dynamic Graph 4 Graph Transformer Network for Continuous-time Dynamic Graph 4.1 Node Sampling 4.2 Graph Transformer Network 4.3 Output Layer 5 Experiments 5.1 Datasets 5.2 Baseline 5.3 Performance 5.4 Discussion 6 Conclusion References ICDT: Incremental Context Guided Deliberation Transformer for Image Captioning 1 Introduction 2 Related Work 2.1 Image Encoding Over Different Features 2.2 Deliberation-Motivated Methods 3 Methodology 3.1 Problem Statement 3.2 Incremental Context Encoder 3.3 Deliberation Decoder 3.4 Training Details 4 Experiments 4.1 Experimental Settings 4.2 Quantitative Analysis 4.3 Ablation Study 4.4 Qualitative Analysis 5 Conclusion References Semantic-Adversarial Graph Convolutional Network for Zero-Shot Cross-Modal Retrieval 1 Introduction 2 Related Work 3 Proposed Method 3.1 Problem Definition 3.2 SAGCN 4 Experiment 4.1 Experimental Setting 4.2 Overall Results 4.3 Further Analysis 5 Conclusion References DAST: Depth-Aware Assessment and Synthesis Transformer for RGB-D Salient Object Detection 1 Introduction 2 Related Work 3 Methodology 3.1 Overview 3.2 Two-Stream Swin Transformer Encoder 3.3 Depth-Aware Assessment and Synthesis (DAS) 3.4 Decoder with Feature Aggregation (FA) 3.5 Loss Function 4 Experiments 4.1 Experimental Setting 4.2 Comparison with SOTA Methods 4.3 Ablation Studies 5 Conclusion References A Vehicle Re-ID Algorithm Based on Channel Correlation Self-attention and Lstm Local Information Loss 1 Introduction 2 Related Work 3 Proposed Method 3.1 Architecture of Our Proposed Method 3.2 CCSAM 3.3 LstmLocal Loss 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Evaluation 4.4 Comparison with Related Methods 4.5 Ablation Study 5 Conclusion References A Self-supervised Graph Autoencoder with Barlow Twins 1 Introduction 2 Related Work 2.1 Graph Autoencoder Models 2.2 Graph Contrastive Learning 2.3 Barlow Twins 3 The Proposed Model 3.1 Overall Framework 3.2 Data Augmentations 3.3 Multi-view Graph Auto-Encoder 3.4 Loss Function 4 Experiment 4.1 Datasets 4.2 Implementation 4.3 Comparison Methods 4.4 Experimental Results 4.5 Random Attacks on Edges 5 Conclusion References Few-Shot Image Classification Method Based on Fusion of Important Features of Different Scales 1 Introduction 2 Related Work 3 MSIFA Method 3.1 Problem Definition 3.2 Feature Extraction Module 3.3 Multi-scale Feature Generation Module 3.4 Self-attention Feature Aggregation Module 4 Experimental Analysis 4.1 Dataset and Evaluation Metrics 4.2 Experimental Setup and Comparison Methods 4.3 Experimental Results and Analysis 4.4 Ablation Experiment 5 Conclusion References Group Residual Dense Block for Key-Point Detector with One-Level Feature 1 Introduction 2 The Proposed Method 2.1 Backbone Network and Detection Head 2.2 U-Fusion Module with Proposed g-RDB 2.3 Detection Missions 3 Experiments and Results 3.1 Training Details 3.2 One-Level vs. Multi-Scale 3.3 Grouped Convolution vs. Depth-Wise Separable Convolution 3.4 With or Without Soft-NKS 3.5 Comparison Results on VOC2007 3.6 Comparison Results on COCO2017 4 Conclusion References Author Index