Sentiment Analysis in the Medical Domain

Contents .- Part I Landscape of medical sentiment.- 1 What is special about medical sentiment analysis?.- 1.1 Overview.- 1.2 Opinion definition.- 1.3 Definition of medical sentiment.- 2 Use cases of medical sentiment analysis.- 2.1 Sentiment analysis in mental health.- 2.2 Outcome and quality assessment of healthcare services and technologies.- 2.2.1 Analysis of patient questionnaires.- 2.2.2 Clinical outcome analysis.- 2.2.3 Social media as mirror of service quality.- 2.3 Sentiment analysis for clinical risk prediction.- 2.4 Sentiment analysis for public health.- 2.5 Sentiment analysis for pharmacovigilance.- 2.6 Sentiment and emotion analysis in health-related conversational agents.- Part II Resources and challenges.- 3 Medical social media and its characteristics.- 3.1 Characteristics of medical social media data.- 3.2 Twitter.- 3.3 User reviews.- 3.4 Forums.- 4 Clinical narratives and their characteristics.- 4.1 Linguistic characteristics of clinical narratives.- 4.2 Clinicalnarratives.- .ix x Contents.- 5 Other data sources.- 5.1 User statements from interaction with intelligent agents.- 5.2 Other sources.- 6 Datasets for medical sentiment analysis.- 6.1 The burden of available datasets.- 6.2 MIMIC databases.- 6.3 i2B2 dataset.- 6.4 TREC dataset.- 6.5 eDiseases dataset.- .6.6 Multimodal Sentiment Analysis Challenge (MuSe).- 6.7 General domain datasets.- 7 Lexical resources for medical sentiment analysis.- 7.1 LIWC.- 7.2 SentiWordNet and its derivations.- 7.3 AFINN.- 7.4 EmoLex.- 7.5 WordNet Affect.- 7.6 WordNet for Medical Events.- 7.7 Other sentiment lexicons.- 7.8 Ontologies and biomedical vocabularies.- .Part III Solutions.- 8 Levels and tasks of sentiment analysis.- 8.1 Level of analysis.- 8.1.1 Document-level sentiment analysis.- 8.1.2 Sentence-level sentiment analysis..- 8.1.3 Aspect-level sentiment analysis..- 8.2 Tasks within medical sentiment analysis..- 8.2.1 Subjectivity analysis..- 8.2.2 Polarity analysis..- 8.2.3 Intensity classification. .- 8.2.4 Emotion recognition..- 9 Document pre-processing.- 9.1 Overview.- 9.2 Data collection and preparation.- 9.3 Text normalisation. .- 9.4 Feature extraction. .- 9.4.1 Bag of words.- 9.4.2 Distributed representation.- 9.5 Feature selection. . .- 9.6 Topic detection..- Contents xi .- Lexicon-based medical sentiment analysis..- 1 Overview on lexicon-based approaches..- 2 Approaches to lexicon generation.- achine learning-based sentiment analysis approaches.- .1 Unsupervised learning approaches . .- .1.1 Partition methods.- 1.2 Hierarchical clustering methods..- 1.2 Supervised approaches .- .2.1 Linear approaches .- .2.2 Probabilistic approaches. .- 2.3 Rule-based classifier.- .2.4 Decision tree classifier. .- .3 Semi-supervised approaches. . .- .4 Deep learning approaches .- .4.1 Deep neural networks (DNN).- .4.2 Convolutional neural networks (CNN).- .4.3 Long short-term memory (LSTM)..- 11.5 Hybrid approaches.- 11.6 Concluding remarks.- 12 Sentiment analysis tools.- 12.1 Sentiment: Sentiment Analysis Tool..- 12.2 TextBlob.- 12.3 Pattern for Python..- 12.4 Valence Aware Dictionary and Sentiment Reasoner (VADER).- 12.5 TensiStrength.- 12.6 LIWC83.- 12.7 Other tools.- 13 Case studies.- 13.1 Learning about suicidal ideation.- 13.1.1 The problem.- 13.1.2 Solution overview.- 13.1.3 Methods and procedures.- 13.2 Predicting the psychiatric readmission risk.- 13.2.1 The problem.- 13.2.2 Solution overview.- 13.2.3 Methods and procedures.- .13.3 Generating a corpus for clinical sentiment analysis.- 13.3.1 The problem.- 13.3.2 Solution overview.- 13.3.3 Methods and procedures. .- 13.4 Conversational agent with emotion recognition.- 13.4.1 The problem.- xii Contents.- 13.4.2 Solution overview.- 13.4.3 Methods and procedures. .- 13.5 Surveillance of public opinions in times of pandemics .- 13.5.1 The problem.- 13.5.2 Solution overview.- 13.5.3 Methods and procedures. .- 13.6 Providing quality information about hospitals.- 13.6.1 The problem.- 13.6.2 Solution overview.- 13.6.3 Methods and procedures. .- Part IV Future.- 14 Medical sentiment analysis - Quo vadis?.- 14.1 SWOT strategy..- 14.2 Strengths.- 14.3 Weaknesses..- 14.4 Opportunities.- 14.5 Threats101 15 Open challenges related to language. .- 15.1 Specific language phenomena hampering sentiment analysis. . .- 15.1.1 Negations .- 15.1.2 Valence shifters .- 15.1.3 Paraphrasing, sarcasm and irony..- 15.1.4 Comparative sentences. .- 15.1.5 Coordination structures.- 15.1.6 Word ambiguity..- 15.2 Evolution of language.- 16 Responsible sentiment analysis in healthcare..- 16.1 Ethical principles applied to medical sentiment analysis.- 16.2 Respect for autonomy.- 16.3 Beneficience and non-maleficience .- 16.4 Justice.- 16.5 Explicability and trust.- 16.6 Concluding remarks.- 17 Explainable sentiment analysis..- 17.1 Definition and need for XAI. . .- 17.2 Explainable AI methods.- 17.3 Applications of XAI to medical sentiment analysis.- Contents xiii 18 The future of medical sentiment analysis.- 18.1 Current research gaps in medical sentiment analysis.- 18.2 Towards domain-specific resources: Lexicons and datasets..- 18.3 Addressing domain-specific challenges and increasing accuracy. .- 18.4 Towards understandable and ethical sentiment analysis..- 18.5 Demonstrate the benefit for patient care..- 18.6 Concluding remarks.- References.- Glossary..- Index.