Computational Intelligence Applications for Text and Sentiment Data Analysis

Sentiment Analysis (SA) has emerged as one of the fastest growing research trends in the last few years as exponential numbers of global internet users are expressing their opinions through various social media platforms across a wide range of issues. Emotion and polarity prediction, from customer feedback through various social media such as Facebook, Twitter, etc., is an important emerging subfield of predictive modelling. Most recently, many big companies have been using various computational intelligence algorithms to understand customers' attitudes towards their products and in order to successfully run their businesses. In this way, Sentiment Analysis has emerged as a critical tool in decision making because social media platforms are used as the most preferred medium to record such issues. Computational Intelligence Applications for Text and Sentiment Data Analysis explores the most recent advances in text information processing and data analysis technologies, specifically focusing on sentiment analysis from multi-faceted data. It investigates a wide range of challenges involved in the accurate analysis of online sentiments, including how to i) identify subjective information from text i.e. exclusion of 'neutral' or 'factual' comments that do not carry sentiment information, ii) identify sentiment polarity, and iii) domain dependency. Spam and fake news detection, short abbreviation, sarcasm, word negation, and a lot of word ambiguity are also explored. Further chapters look at the difficult process of extracting sentiment from different multimodal information (audio, video and text), semantic concepts. In each chapter, authors explore how computational intelligence (CI) techniques, such as deep learning, convolutional neural network, fuzzy and rough set, global optimizers, and hybrid machine learning techniques, play an important role in solving the inherent problems of sentiment analysis applications. Cover Contents About the editors List of contributors Preface 1 Sentiment analysis and computational intelligence 1.1 Introduction 1.2 Themes and challenges 1.3 Goals 1.4 Recent advances in sentiment-data analysis 1.4.1 Industrial and academic developments 1.5 Conclusions References 2 Natural language processing and sentiment analysis: perspectives from computational intelligence 2.1 Introduction 2.2 Related work 2.3 Popular emotion models in computational analysis 2.3.1 Russell’s circumplex model of emotions 2.3.2 Plutchik’s wheel of emotions 2.3.3 Ekman’s basic emotions 2.3.4 Hourglass of emotions 2.4 Dataset description 2.4.1 Corpus preparation 2.4.2 Descriptive statistics of the corpus 2.4.3 Generation of sentiment labels using weak supervision 2.5 Proposed methodology 2.5.1 Word-embedding representation 2.5.2 Base models 2.5.2.1 Convolutional Neural Network (CNN_STL) 2.5.2.2 Bidirectional Long Short-Term Memory (BiLSTM_STL) 2.5.2.3 Bidirectional Gated Recurrent Unit (Bi-GRU_STL) 2.5.3 Attention 2.5.4 Loss function 2.5.5 Multitask framework 2.6 Baselines 2.7 Experiments, results, and analysis 2.7.1 Experimental setting 2.7.2 Evaluation metrics 2.7.3 Experimental results and discussion 2.7.4 Error analysis 2.8 Conclusion Acknowledgments References 3 Applications and challenges of SA in real-life scenarios 3.1 Health 3.1.1 Challenges 3.1.2 Applications 3.2 Social policy 3.2.1 Applications in social policy 3.2.2 Challenges in social policy 3.3 E-commerce / industry 3.3.1 Applications in e-commerce 3.3.2 Challenges in e-commerce 3.4 Digital humanities 3.4.1 Applications 3.5 Other research areas 3.5.1 Applications to other research areas 3.5.2 Challenges for SA research in other areas 3.6 Summary References 4 Emotions of students from online and offline texts 4.1 Introduction 4.2 State-of-the-art 4.3 Some basic concepts 4.3.1 Natural language processing (NLP) 4.3.2 Image processing 4.3.3 Graphology-oriented feature extraction 4.3.4 Rule-based classifier 4.4 Methodology 4.4.1 Raw-data collection 4.4.1.1 Phase 1: data collection of social-media messages 4.4.1.2 Phase 2: data collection of signature images 4.4.2 Phase 1: text-message analysis 4.4.2.1 Data pre-processing 4.4.2.2 Applying lexical analysis 4.4.2.3 Generating mood-oriented comments 4.4.3 Phase 2: signature image analysis 4.4.3.1 Pre-processing: handwritten digitized signatures 4.4.3.2 Feature-extraction algorithms from signature images 4.4.3.3 Applying rule-based classifiers 4.5 Result and performance analysis 4.5.1 Phase 1: computerized analysis results on message dataset 4.5.1.1 Sample score graphs for Participants #1, and #11 to #14 4.5.2 Phase 2: computerized analysis results on signature dataset 4.5.3 Final output 4.6 Conclusion and future scope Acknowledgments References 5 Online social-network sensing models 5.1 Introduction 5.2 Importance of sensing in online social networks 5.2.1 Why social sensing 5.2.2 Areas of applications 5.3 Sensing modalities 5.3.1 Network analysis 5.3.1.1 Network properties 5.3.1.1.1 Density, distance, diameter 5.3.1.1.2 Degree distribution 5.3.1.1.3 Power-law exponent 5.3.1.1.4 Clustering coefficient 5.3.1.2 Global clustering coefficient 5.3.1.3 Local clustering coefficient 5.3.1.4 Average local clustering coefficient 5.3.1.4.1 Centrality 5.3.1.4.2 Homophily 5.3.1.4.3 Assortativity 5.3.1.5 Propagation models 5.3.2 Text analysis 5.3.2.1 Collection of information 5.3.2.2 Information retrieval 5.3.2.3 Text-mining 5.3.2.4 Sentence clustering 5.3.2.5 Sentiment analysis 5.3.3 Smartphone-sensor data analysis 5.3.3.1 Addiction 5.3.3.2 Cognitive absorption 5.4 Applications 5.4.1 Link prediction 5.4.2 Event prediction 5.4.3 Marketing 5.4.4 Influence maximization 5.4.5 Rumor blocking 5.4.6 Mental health 5.4.7 Social awareness 5.5 Conclusion References 6 Learning sentiment analysis with word embeddings 6.1 Introduction 6.2 Word-embedding representations 6.2.1 Word2Vec 6.2.2 GloVe 6.3 Contextual word representations 6.3.1 ELMo (deep contextualized word representations) 6.3.2 BERT (pre-training of deep bidirectional transformers for language understanding) 6.4 Sentiment-analysis definition 6.5 Model architecture 6.5.1 MoCE architecture 6.5.2 Multigrained gcForest architecture 6.6 Experimental setup 6.6.1 Dataset description 6.6.2 Feature extraction 6.6.3 Training strategy 6.7 Results & discussion 6.7.1 Evaluation of embeddings using MoCE 6.7.2 Polarity identification using gcForest 6.8 Conclusion References 7 An annotation system of a medical corpus using sentiment-based models for summarization applications 7.1 Introduction 7.2 Background work 7.3 Dataset preparation 7.4 Proposed annotation system 7.4.1 Medical-concept identification 7.4.2 Sentiment identification 7.5 Sentiment-based summarization 7.6 Results analysis 7.7 Conclusions References 8 A comparative study of a novel approach with baseline attributes leading to sentiment analysis of Covid-19 tweets 8.1 Introduction 8.1.1 Motivation and objective 8.2 Related study 8.2.1 Groundwork of sentiment analysis 8.2.2 Sentiment analysis from the web and social media 8.2.3 Sentiment analysis in healthcare 8.2.4 Sentiment analysis on Covid-19 8.3 Comparisons of methodology on recent Covid-19-based NLP research 8.4 Proposed contribution 8.4.1 Preparing Covid-19 dataset 8.4.2 Data pre-processing 8.4.3 Attribute I: word-trend detection using n-gram 8.4.4 Attribute II: Covid-19-specific word identification 8.4.5 Sentiment classification 8.4.6 Sentiment modeling using bidirectional long short-term model 8.5 Benchmark comparison of recent NLP-based experiments on Covid-19 8.6 Discussion and future scope 8.7 Conclusion References 9 Sentiment-aware design of human–computer interactions 9.1 Introduction and definitions 9.2 Relationship between HCI and sentiment analysis 9.3 Emotional intelligence 9.4 Sentiment-aware conversational systems 9.5 Methodological approaches 9.6 Future directions 9.7 Ethical considerations References 10 Towards the analyzing of social-media data to assess the impact of long lockdowns on human psychology due to the Covid-19... 10.1 Introduction 10.1.1 Major contribution of the chapter 10.1.2 Organization of the chapter 10.2 Related work 10.3 Data layout 10.4 Our approach 10.5 Algorithm 10.5.1 Algorithm for analyzing the impact of long lockdowns on human psychology 10.5.2 Discussion 10.5.3 Performance analysis 10.5.4 A few findings 10.6 Conclusion and future scope References 11 Conclusion 11.1 Introduction 11.2 Discussion 11.3 Concluding remarks Index Computational Intelligence Applications for Text and Sentiment Data Analysis explores the most recent advances in text information processing and data analysis technologies, specifically focusing on sentiment analysis from multifaceted data. The book investigates a wide range of challenges involved in the accurate analysis of online sentiments, including how to i) identify subjective information from text, i.e., exclusion of ‘neutral’ or ‘factual’ comments that do not carry sentiment information, ii) identify sentiment polarity, and iii) domain dependency. Spam and fake news detection, short abbreviation, sarcasm, word negation, and a lot of word ambiguity are also explored. Further chapters look at the difficult process of extracting sentiment from different multimodal information (audio, video and text), semantic concepts. In each chapter, the book's authors explore how computational intelligence (CI) techniques, such as deep learning, convolutional neural network, fuzzy and rough set, global optimizers, and hybrid machine learning techniques play an important role in solving the inherent problems of sentiment analysis applications.