[Algorithms for Intelligent Systems] Deep Learning-Based Approaches for Sentiment Analysis ||

"This book covers deep-learning-based approaches for sentiment analysis, a relatively new, but fast-growing research area, which has significantly changed in the past few years. The book presents a collection of state-of-the-art approaches, focusing on the best-performing, cutting-edge solutions for the most common and difficult challenges faced in sentiment analysis research. Providing detailed explanations of the methodologies, the book is a valuable resource for researchers as well as newcomers to the field"--Back cover Preface 6 Contents 9 About the Editors 11 Application of Deep Learning Approaches for Sentiment Analysis 13 1 Introduction 14 2 Taxonomy of Sentiment Analysis 14 2.1 Sentiment Analysis, Polarity, and Output 16 2.2 Levels of Sentiment Analysis 16 2.3 Domain Applicability, Training, and Testing Strategy 17 2.4 Language Support 18 2.5 Evaluation Measures 18 3 Text Representation for Sentiment Analysis 18 3.1 Embedded Vectors 18 3.2 Strategy of Initializing the Embedded Vectors 21 3.3 Enhancing the Embedded Vectors 21 3.4 Approximation Methods 22 3.5 Sampling-Based Approaches 22 3.6 Softmax-Based Approaches 23 4 Deep Learning Approaches for Sentiment Analysis 23 5 Evaluation Metrics for Sentiment Analysis 30 6 Benchmarked Datasets and Tools 33 7 Conclusion 35 References 38 Recent Trends and Advances in Deep Learning-Based Sentiment Analysis 44 1 Introduction 45 2 Related Work 46 3 Machine Learning Approaches for Sentiment Analysis 46 4 Study Rationale 48 5 Deep Learning Architectures 49 5.1 Convolutional Neural Networks 49 5.2 Recurrent Neural Networks 50 5.3 Bi-directional Recurrent Neural Network 50 6 Long Short-Term Memory (LSTMs) 51 7 Gated Recurrent Units (GRUs) 53 8 Attention Mechanism 54 9 Research Methodology 55 10 Approach to Sentiment Analysis Task Categorization 55 11 Coarse-Grain Sentiment Analysis 56 12 Fine-Grain Sentiment Analysis 58 13 Cross-Domain Sentiment Analysis 61 14 Conclusion and Survey Highlights 62 References 62 Deep Learning Adaptation with Word Embeddings for Sentiment Analysis on Online Course Reviews 68 1 Introduction 69 2 State of the Art 71 2.1 Sentiment Analysis in E-Learning Systems 71 2.2 Deep Learning for Sentiment Analysis 72 2.3 Word Embeddings for Sentiment Analysis 73 3 Word Embedding Representations for Text Mining 74 3.1 Word2Vec 75 3.2 GloVe 75 3.3 FastText 75 3.4 Intel 76 4 Deep Learning Components for Text Mining 76 4.1 Feed-Forward Neural Network (FNN) 76 4.2 Recurrent Neural Network (RNN) 77 4.3 Long Short-Term Memory (LSTM) Network 78 4.4 Convolutional Neural Network (CNN) 78 4.5 Normalization Layer (NL) 79 4.6 Attention Layer (AL) 79 4.7 Other Layers 79 5 Our Sentiment Predictor for E-Learning Reviews 80 5.1 Review Splitting 80 5.2 Word Embedding Modeling 81 5.3 Review Vectorization 82 5.4 Sentiment Model Definition 83 5.5 Sentiment Model Training and Prediction 85 6 Experimental Evaluation 85 6.1 Dataset 85 6.2 Baselines 85 6.3 Metrics 86 6.4 Deep Neural Network Model Regressor Performance 87 6.5 Contextual Word Embeddings Performance 87 7 Conclusions, Open Challenges, and Future Directions 90 References 92 Toxic Comment Detection in Online Discussions 95 1 Online Discussions and Toxic Comments 95 1.1 News Platforms and Other Online Discussions Forums 96 1.2 Classes of Toxicity 97 2 Deep Learning for Toxic Comment Classification 99 2.1 Comment Datasets for Supervised Learning 99 2.2 Neural Network Architectures 101 3 From Binary to Fine-Grained Classification 104 3.1 Why Is It a Hard Problem? 104 3.2 Transfer Learning 106 3.3 Explanations 107 4 Real-World Applications 109 4.1 Semi-automated Comment Moderation 110 4.2 Troll Detection 111 5 Current Limitations and Future Trends 112 5.1 Misclassification of Comments 112 5.2 Research Directions 114 6 Conclusions 115 References 116 Aspect-Based Sentiment Analysis of Financial Headlines and Microblogs 120 1 Introduction 121 2 Related Work 123 3 State-of-the-Art Models 124 3.1 ALA Model 124 3.2 IIIT Delhi Model 125 4 Our Methodology 126 4.1 Features 127 5 Aspect Classification Models 129 5.1 Models 129 5.2 Classification Model Training 131 6 Sentiment Models 131 6.1 Models 131 6.2 Sentiment Model Training 136 7 Evaluation 137 7.1 Data Set 137 7.2 Data Augmentation 138 7.3 Data Pre-processing 138 7.4 Metrics 140 7.5 Results 141 8 Conclusion and Future Work 142 References 143 Deep Learning-Based Frameworks for Aspect-Based Sentiment Analysis 147 1 Introduction 147 2 Problem Formulation 150 2.1 Aspect-Term Extraction 150 2.2 Aspect-Category Detection 150 3 Observation/Assumption in ABSA 150 4 Input Representation 151 5 Concepts Related to Deep Learning 152 5.1 Word-Embeddings 152 5.2 Long Short-Term Memory (LSTM) 153 5.3 Bi-directional Long Short-Term Memory (Bi-LSTM) 154 5.4 RNN with Attention 155 5.5 Convolution Neutral Network (CNN) 157 6 Deep Learning Architectures Used in ABSA 158 6.1 Sentiment Analysis 158 6.2 Aspect-Term Extraction 158 6.3 Aspect-Category Extraction 159 6.4 Aspect-Based Sentiment Detection 160 7 Conclusion 164 References 164 Transfer Learning for Detecting Hateful Sentiments in Code Switched Language 167 1 Introduction 168 1.1 Hate Speech Problem 168 1.2 Code Switched and Code Mixed Languages 169 1.3 Challenges in Code Switched and Code Mixed Languages 170 1.4 Deep Learning 170 1.5 Overview 170 2 Background and Related Work 171 2.1 Language Identification 172 2.2 POS Tagging 172 2.3 Named Entity Recognition 174 2.4 Sentiment Analysis 175 3 Dataset and Evaluation 177 3.1 HOT Dataset 178 3.2 Bohra et al. bohra2018dataset dataset 179 3.3 HEOT Dataset 179 3.4 Davidson Dataset 181 4 Methodology 181 4.1 SVM and Random Forest 181 4.2 Ternary Trans-CNN Model 183 4.3 LSTM-Based Model 186 4.4 MIMCT Model 188 5 Results 191 5.1 SVM and Random Forest Classifier 191 5.2 Ternary Trans-CNN Model 192 5.3 LSTM Model with Transfer Learning 192 5.4 MIMCT Model 194 6 Conclusion 195 7 Future Work 195 References 197 Multilingual Sentiment Analysis 201 1 Introduction 202 1.1 Low Resource Language 202 1.2 Challenges of Sentiment Analysis 203 1.3 Deep Learning 204 2 Literature Survey 205 2.1 High Resource Languages 205 2.2 Lexicon-Based Approaches 206 2.3 Traditional Machine Learning-Based Approaches 207 2.4 Low Resource Languages 208 3 Word Embeddings for Sentiment Analysis 209 3.1 Refining Word Embeddings for Sentiment Analysis 210 3.2 Improving Word Embedding Coverage in Low Resource Languages 213 4 Deep Learning Techniques for Multilingual Sentiment Analysis 216 4.1 Convolutional Neural Networks 217 4.2 Recurrent Neural Networks 220 4.3 Autoencoders 228 4.4 Bilingual Constrained Recursive Autoencoders 230 4.5 AROMA 232 4.6 Siamese Neural Networks 235 5 Discussion 237 6 Conclusion 241 References 241 Sarcasm Detection Using Deep Learning-Based Techniques 245 1 Introduction 245 2 Related Work 248 3 Grice’s Maxims 251 4 Challenges in Sarcasm Detection 255 5 Dataset Description 256 6 Feature Description 258 7 Process Outline 263 8 Models Used 263 9 Experiments and Results 264 10 Future Scope 265 References 265 Deep Learning Approaches for Speech Emotion Recognition 267 1 Introduction 268 2 Feature Extraction 269 3 Feature Selection 271 4 Classical Approaches 271 4.1 Speaker-Dependent SER 272 4.2 Speaker-Independent SER 273 4.3 Other Models 277 5 Deep Learning Approaches 278 6 System Overview 281 6.1 Classical Approach for SER 281 6.2 Deep Learning Approaches for SER 282 6.3 Critical Comparision 287 7 Evaluation 288 7.1 Dataset Description 288 7.2 Original Results 288 7.3 Results Obtained 290 8 Comparison of Existing Approaches 290 9 Conclusions 291 References 291 Bidirectional Long Short-Term Memory-Based Spatio-Temporal in Community Question Answering 298 1 Introduction 299 2 Related Works 301 3 Methodology 304 3.1 Preprocessing Steps 304 3.2 Best Answer Prediction 305 4 Experimental Setup: Answer Classification 310 5 Experiment II: Answer Ranking 313 6 Conclusion 315 References 315 Comparing Deep Neural Networks to Traditional Models for Sentiment Analysis in Turkish Language 318 1 Introduction 319 2 Methodology 320 3 Experimental Setup and Results 321 3.1 Dataset 321 3.2 Traditional BOW Approach 321 3.3 Deep Learning Architecture 323 4 Conclusion 325 References 325