Title Page
Abstract
Contents
Chapter 1. Introduction 10
1.1. Backgrounds of Research 10
1.2. Objectives of the Research 11
1.3. Contribution of the Research 12
1.4. Structure of the Research 12
Bibliography 15
Chapter 2. Crisis communication and Methodologies for Measuring Sentiment of the Public 16
2.1. Crisis Communication in Disasters 16
2.2. Corpus and Corpus Linguistics for Analyzing Contents of Crisis Communication 17
2.2.1. Corpus and Corpus Linguistics 17
2.2.2. Crisis Communication using Twitter 19
2.3. Sentiment Analysis for Crisis Communication using Corpus 22
2.3.1. Reasons for Sentiment Analysis 22
2.3.2. Utility of Twitter Corpus as a Data for Crisis Communication Analysis 23
2.4. Methodologies for Sentiment Analysis in Disasters 24
2.4.1. Review of Related Research 24
2.4.2. Topic models as a Tools for analyzing Twitter Data 26
2.4.3. Topic Model (LDA) 26
2.4.4. Topic Model (TFIDF) 27
2.5. Conclusion 28
Bibliography 29
Chapter 3. Analysis of Tweets in Disaster 33
3.1. Introduction 33
3.2. The Great East Japan Earthquake and The Fukushima Daiichi Nuclear Disaster 34
3.3. Tweets as an utterance of the public and its Transmission in Twitter 34
3.3.1. Data 34
3.3.2. Tweeting by the public Related to the Disaster 37
3.3.3. Transmission of Tweets by Retweet Related to the Disaster 39
3.4. Disaster Information Distribution by Government Agencies 41
3.4.1. The Contents of Disaster Information Distribution 41
3.4.2. Disaster Information Transmission by Retweets 47
3.5. Conclusion 47
Bibliography 49
Chapter 4. Estimating Concerns of the Public with Latent Dirichlet Allocation 50
4.1. Introduction 50
4.2. Basic Idea 51
4.2.1. Concerns of the public in disasters 51
4.2.2. LDA Topic Model for Identifying Concerns of the Public 52
4.3. Data 53
4.4. The Methodology and The Application 54
4.4.1. Latent Variable Topic Model (Latent Dirichlet Allocation) 54
4.4.2. Model Inference 56
4.4.3. Application for clarifying the concerns of the publics 60
4.5. A Changing Concerns of the Public 62
4.5.1. Model Selection 62
4.5.2. Fitting the LDA Model to the Twitter Data Set Using 30 Topics 62
4.6. Implication 74
4.7. Conclusion 74
Bibliography 76
Chapter 5. Measurement of Disaster Anxiety of the public 78
5.1. Introduction 78
5.2. Basic Idea 79
5.2.1. Anxiety as risk perception of the public 79
5.2.2. Utility of the Twitter corpus as data for evaluating public sentiment 81
5.3. Data 83
5.4. The Methodology and The Application 84
5.5. Measuring Anxiety using Anxiety Index 87
5.6. Implication 91
5.7. Conclusion 92
Bibliography 93
Chapter 6. Conclusions and Future Research 95
6.1. Conclusions 95
6.2. Topics for Future Research 97
Bibliography 99
Table 3.1. The Number of Tweets 36
Table 3.2. The Total Number of Tweets containing words Radiation(HOUSYA), Earthquake(JISHIN) and Tsunami 38
Table 3.3. The Proportion in Tweets including Word 'Radiation (HOUSYA)' 39
Table 3.4. The 10 most Retweeted Twitter Account and the The number of Times their... 40
Table 3.5. The number of Tweets provided by Government Agency 42
Table 3.6. The Contents of Information provided by Government Agencies 43
Table 3.6. The Contents of Information provided by Government Agencies 44
Table 3.6. The Contents of Information provided by Government Agencies 45
Table 3.7. The Contents of Information provided by Government Agencies of The Dis-... 46
Table 3.8. The Number of Retweets of Government Agencies's Tweets 48
Table 4.1. The Outline of Sample Data 53
Table 4.2. The 10 Highest probability words for each of 30 Topics 66
Table 4.3. Topic Changing from March 11 to 17 (not frequently changing topics) 71
Table 5.1. The Outline of Sample Data 83
Table 5.2. 40 Highest Ranked Negative Co-Occurrence Frequency Words 89
Table 5.3. The Time Series Variation of Co-Occurrence Frequency with 'HOUSYA... 90
Table 5.4. The Time Series Variation of Co-Occurrence Frequency with 'HOUSYA... 91
Figure 1.1. Research Process and Framework 13
Figure 2.1. Crisis Communication in Disaster using Twitter 20
Figure 2.2. The Concept of Formation of Collective Actions 23
Figure 3.1. Twitter Data comprising tweet IDs, user IDs, time and tweet contents 35
Figure 3.2. Data provided by The Great East Japan Earthquake Big Data Workshop... 36
Figure 3.3. The Time Series of the Quantity of Tweets Containing Words Radia-... 37
Figure 4.1. Graphical model representation of LDA 54
Figure 4.2. (Left) Graphical model representaion of LDA. (Right) Graphical model... 57
Figure 4.3. A variational inference algorithm for LDA 59
Figure 4.4. Outline of Application 61
Figure 4.5. Perplexities of the test data for the models fitted with LDA. Each line... 63
Figure 4.6. Estimated α values for the models fitted. Each line corresponds to one of... 64
Figure 4.7. Topic Changing from March 11 to 17 (frequently changing topics) 70
Figure 5.1. Time Series of Variation of Volume of Queries (Google Trends) 79
Figure 5.2. Time Series of Variation of Volume of Queries (Google Trends) 80
Figure 5.3. Time Series of Variation of Volume of Tweets including HOUSYA (radiation) 84
Figure 5.4. Outline of the Application 85
Figure 5.5. The List of Words and Semantic Orientations for Japanese 86
Figure 5.6. Time Series Variation of Anxiety 88