Combating On​line Ho​stile Posts in ​Regional Languages During Emerge​ncy Si​tuation: First International Workshop, CONSTRAINT 2021 Collocated with AAAI 2021 Virtual Event, February 8, 2021, Revised Selected Papers

This book constitutes selected and revised papers from the First International Workshop on Combating On​line Ho​st​ile Posts in ​Regional Languages dur​ing Emerge​ncy Si​tuation, CONSTRAINT 2021, Collocated with AAAI 2021, held as virtual event, in February 2021. The 14 full papers and 9 short papers presented were thoroughly reviewed and selected from 62 qualified submissions. The papers present interdisciplinary approaches on multilingual social media analytics and non-conventional ways of combating online hostile posts. Preface Organization Contents Identifying Offensive Content in Social Media Posts 1 Introduction 2 Related Work 3 Dataset 4 Methodology 4.1 Feature Extraction and Machine Learning 4.2 Sentence Embedding and Machine Learning 4.3 Fine-Tuning BERT and DistilBERT 5 Results and Evaluation 5.1 Hyperparameter Tuning 6 Conclusion References Identification and Classification of Textual Aggression in Social Media: Resource Creation and Evaluation 1 Introduction 2 Related Work 3 Task Definition 3.1 Level A: Aggressive Text Identification 3.2 Level B: Classification of Aggressive Text 4 Aggressive Text Corpus 4.1 Corpora Development 4.2 Corpora Analysis 5 Methodology 6 Experiments and Result Analysis 6.1 Error Analysis 7 Conclusion References Fighting an Infodemic: COVID-19 Fake News Dataset 1 Introduction 2 Related Work 3 Dataset Development 3.1 Collection and Annotation 3.2 Fake News 3.3 Real News 3.4 Dataset Statistics 4 Baselines and Results 5 Conclusion and Future Work References Revealing the Blackmarket Retweet Game: A Hybrid Approach 1 Introduction 2 Dataset 3 A Hybrid Detection Framework 3.1 Indicators 4 Experiments 4.1 Classification 4.2 Quantitative Analysis 4.3 Graph-Based Analysis 5 Conclusion and Future Work References Overview of CONSTRAINT 2021 Shared Tasks: Detecting English COVID-19 Fake News and Hindi Hostile Posts 1 Introduction 2 Related Work 3 COVID-19 Fake News Detection in English 4 Hostile Post Detection in Hindi 5 Participation and Top Performing Systems 5.1 Winning Systems 5.2 Interesting Systems 6 Results 7 Conclusion and Future Work References LaDiff ULMFiT: A Layer Differentiated Training Approach for ULMFiT 1 Introduction 2 Overview 2.1 Task Description and Dataset 2.2 Preprocesing 3 Model Description 3.1 Layer Differentiated ULMFiT Training 3.2 Customized RoBERTa 3.3 Random Forest Classifiers and Logistic Regression 4 Results 5 Conclusions References Extracting Latent Information from Datasets in CONSTRAINT 2021 Shared Task 1 Introduction 2 Related Work 3 Method 3.1 English Fake News Detection Task 3.2 Hindi Hostile Post Detection Task 4 Experiment 4.1 Dataset 4.2 Baseline Model 4.3 Optimization Approach 5 Conclusion References Fake News and Hostile Posts Detection Using an Ensemble Learning Model 1 Introduction 2 Related Work 3 Datasets 4 Methodology 4.1 BERT 4.2 Ensemble 5 Experiment 6 Results 7 Conclusion References Transformer-Based Language Model Fine-Tuning Methods for COVID-19 Fake News Detection 1 Introduction 2 Related Work 2.1 Text Classification Task with Adversarial Training Methods 2.2 Model Fusion Approaches for Text Classification 2.3 Fake News Detection 3 Methodology 3.1 Problem Definition 3.2 Our Proposed Network 4 Experimental Results 4.1 Experimental Setting 4.2 Performance in Fake News Detection in English 4.3 Ablation Studies for Ro-CT-BERT 5 Conclusions References Tackling the Infodemic: Analysis Using Transformer Based Models 1 Introduction 2 Related Work 3 Dataset Analysis 4 Our Work 4.1 Dataset Collection 4.2 Dataset Cleaning 5 Training 5.1 Hardware Specifications 5.2 Model Descriptions 6 Results 6.1 Performance Scores 6.2 Result Analysis 7 Future Work References Exploring Text-Transformers in AAAI 2021 Shared Task: COVID-19 Fake News Detection in English 1 Introduction 2 Dataset 2.1 Data Source 2.2 Exploratory Data Analysis 3 Methodology 3.1 Text-RNN 3.2 Text-Transformers 4 Experiments 4.1 Experimental Settings 4.2 Training Strategy 4.3 Results 4.4 Analysis 5 Related Work 5.1 Pre-trained Language Models 5.2 K-fold Cross-validation 5.3 Fake News Detection and Categorization 6 Conclusion References g2tmn at Constraint@AAAI2021: Exploiting CT-BERT and Ensembling Learning for COVID-19 Fake News Detection 1 Introduction 2 Related Work 3 Task Definition 4 Dataset 5 Our Approach 5.1 Data Preprocessing 5.2 Models 5.3 Additional Data 5.4 Experimental Settings 6 Results and Discussion 6.1 Comparison of Models for Fake News Detection 6.2 Final Submissions 6.3 Error Analysis 7 Conclusion References Model Generalization on COVID-19 Fake News Detection 1 Introduction 2 Dataset 3 Methodology 3.1 Task and Objective 3.2 Approach 1: Fine-Tuning Pre-trained Transformer Based Language Models with Robust Loss Functions 3.3 Approach 2: Data Noise Cleansing Based on Training Instance Influence 4 Experiment 1: Fine-Tuning LMs with Robust Loss Functions 4.1 Experiment Set-Up 4.2 Experimental Results 5 Experiment 2: Data Cleansing with Influence Calculation 5.1 Experiment Set-Up 5.2 Experiment Result 6 Discussion 6.1 Data Distribution Between Different FakeNews-19 and Tweets-19 Test Sets 6.2 How Did Smaller Data Help for Generalization Ability of the Model? 7 Related Works 8 Conclusion References ECOL: Early Detection of COVID Lies Using Content, Prior Knowledge and Source Information 1 Introduction 2 Related Work 2.1 Textual Content Based Fake News Detection on Social Media 2.2 Extrinsic Features for Determining Truthfulness of Claims 3 ECOL Approach 3.1 Content (C) 3.2 Prior Knowledge (PK) 3.3 Source (S) 4 Experiments 4.1 Dataset 4.2 Baselines 4.3 Models 4.4 Implementation 5 Results and Discussion 6 Conclusion References Evaluating Deep Learning Approaches for Covid19 Fake News Detection 1 Introduction 2 Related Works 3 Architecture Details 3.1 CNN 3.2 LSTM 3.3 Bi-LSTM + Attention 3.4 HAN 3.5 Transformers 4 Experimental Setup 4.1 Dataset Details 4.2 Preprocessing of the Dataset 4.3 Training Details 5 Results and Discussion 6 Conclusion References A Heuristic-Driven Ensemble Framework for COVID-19 Fake News Detection 1 Introduction 2 Related Work 3 Dataset Description 4 Methodology 4.1 Text Preprocessing 4.2 Tokenization 4.3 Backbone Model Architectures 4.4 Ensemble 4.5 Heuristic Post-processing 5 Experiments and Results 5.1 System Description 5.2 Performance of Individual Models 5.3 Performance of Ensemble Models 5.4 Performance of Our Final Approach 5.5 Ablation Study 6 Conclusion References Identification of COVID-19 Related Fake News via Neural Stacking 1 Introduction 2 Related Work 3 Data Description 4 Proposed Method 4.1 Hand Crafted Features 4.2 Latent Semantic Analysis 4.3 Contextual Features 4.4 tax2vec Features 5 Meta Models 5.1 Neural Stacking 5.2 Linear Stacking 6 Experiments and Results 6.1 Train-Development-Test (TDT) Split 6.2 CV Split 6.3 Evaluating Word Features 6.4 Results 7 Conclusion and Further Work References Fake News Detection System Using XLNet Model with Topic Distributions: CONSTRAINT@AAAI2021 Shared Task 1 Introduction 2 Related Work 3 Proposed Method 4 Dataset Description 5 Experiments 5.1 Implementation 5.2 Comparison with Other Methods 5.3 Results and Discussion 6 Error Analysis 7 Conclusion References Coarse and Fine-Grained Hostility Detection in Hindi Posts Using Fine Tuned Multilingual Embeddings 1 Introduction 2 Related Works 3 Methodology 3.1 Coarse-Grained Classification 3.2 Fine-Grained Classification 4 Experimental Setup 5 Results and Discussion 5.1 Coarse-Grained Evaluation 5.2 Fine-Grained Evaluation 5.3 Comparison with Baseline 5.4 Additional Discussions and Analysis 6 Conclusion and Future Works References Hostility Detection in Hindi Leveraging Pre-trained Language Models 1 Introduction 2 Related Work 3 Methodology 3.1 Single Model Multi-label Classification 3.2 Multi-task Classification 3.3 Binary Classification 3.4 Auxiliary Task Based Binary Sub-classification 4 Experiment 4.1 Dataset Description 4.2 Pre-processing 4.3 Experimental Setup 5 Results 6 Error Analysis 7 Conclusion and Future Work References Stacked Embeddings and Multiple Fine-Tuned XLM-RoBERTa Models for Enhanced Hostility Identification 1 Introduction 2 Related Work 3 Task and Dataset Description 4 System Description 4.1 Preprocessing 4.2 Models 4.3 Data Augmentation 5 Results and Discussion 6 Analysis 7 Error Analysis 8 Conclusion and Future Work References Task Adaptive Pretraining of Transformers for Hostility Detection 1 Introduction 2 Dataset 3 Approach 3.1 Preprocessing and Feature Extraction 3.2 Architecture 3.3 Task Adaptive Pretraining 4 Results 5 Experimental Details 6 Conclusion References Divide and Conquer: An Ensemble Approach for Hostile Post Detection in Hindi 1 Introduction 2 Related Work 3 Proposed Methodology 3.1 Model 4 Implementation 4.1 Dataset 4.2 Experiments 4.3 Binary Relevance Considerations 5 Result and Analysis 6 Insights 7 Error Analysis 8 Outcome 9 Future Work References Author Index This book constitutes selected and revised papers from the First International Workshop on Combating On line Ho st ile Posts in Regional Languages dur ing Emerge ncy Si tuation, CONSTRAINT 2021, Collocated with AAAI 2021, held as virtual event, in February 2021. The 14 full papers and 9 short papers presented were thoroughly reviewed and selected from 62 qualified submissions. The papers present interdisciplinary approaches on multilingual social media analytics and non-conventional ways of combating online hostile posts