This paper clarifies the occurrence factors of commuters unable to return home and the returning-home decision-making at the time of the Great East Japan Earthquake by using Twitter data. First, to extract the behavior data from the tweet data, we identify each user’s returning-home behavior using support vector machines. Second, we create non-verbal explanatory factors using geo-tag data and verbal explanatory factors using tweet data. The non-verbal explanatory factors include distance between home and office, time taken in travelling by walking or public transport, etc. On the other hand, the verbal explanatory factors include external and psychological factors. Then, we model users’ returning-home decision-making by using a discrete choice model and clarify the factors quantitatively. Finally, by sensitivity analysis, we show the effects of the existence of emergency evacuation facilities and line of communication.

View full abstract

We analyzed tweets broadcasted until four days after the occurrence of the Great East Japan Earthquake, which are provided by the Project 311. After obtaining a general view from tweets clustering, we created a set of targeted extraction categories from them and constructed a tweet extractor tailored to the target. In a sequence of such processes, improvement of the clustering, which is used to discover the target category for extraction, becomes very important. A method is proposed that utilizes the Singular Value as weights for features, while the well-known conventional use of Singular Value Decomposition is limited to reducing its dimension. In addition, we proposed an evaluation criterion for a human-aided clustering task, and conducted experiments to compare these criteria, including commonly-used ones, with the actual time spent by humans for performing such a task. The experiments show the effectiveness of the proposed weighting method and the competency of our criterion, mainly from the perspective of time efficiency of the task. As for the targeted data-extraction task, which is also a classification problem, some improvement in accuracy is observed although the training process itself involves a weighting mechanism.

View full abstract

In this paper we introduce an information analysis system that automatically acquires from social media like Twitter the kind of vital information that rescue workers or disaster victims need in case of large-scale disasters like earthquakes or tsunami. This system uses question-answering (QA) technology with the aim of helping users get a comprehensive overview of the state of affairs and the various conditions of ongoing rescue efforts in the afflicted areas, and detect potentially unanticipated information from the disaster areas. The system expands the input question with various entailment expressions, and augments the original social media input data by analyzing the mentioned place names in order to handle a wide variety of questions relevant to disaster scenarios. Moreover, we extend this system to address rescue workers’ crucial problem of getting access to relevant information from the disaster areas in times of crisis. To tackle this problem we introduce a mechanism by which rescue workers from NPOs or municipalities can register certain questions for situation assessment in advance, so that when a disaster victim posts some urgent requests for food, medicines or other essentials on Twitter or some other BBS, both information sender and information requester are automatically notified of this. We expect that such a mechanism can safeguard the two-way communication between rescue workers and disaster victims, and ultimately lead to a more effective rescue effort. We evaluate the system on a test set of 300 questions and their answers. For 192 questions whose answers are actually included in our system’s index, we obtained on average 605.8 answers per question, with 51.9% recall and 60.8% precision.

View full abstract

In the early stages of the Great East Japan Earthquake, a vast number of tweets were related to high-urgency rescue requests; however, most of these tweets were buried under many other tweets, including some well-intentioned retweets of the rescue requests. To better handle such a situation, the authors have developed and published a website that automatically lists similar statements to extract rescue requests from Twitter on March 16, 2011. This paper describes not only the technology of the system but also the start of a rescue project #99japan. The project takes particular note of the progress and completion reports of the rescue situations, using this site as sources of rescue information. Note that #99japan originated from a thread of the Japanese textboard 2channel, which was launched by some volunteers within two hours of the disaster’s occurrence.

View full abstract

Although Twitter played an important role in supporting victims of the 2011 Tohoku earthquake and tsunami disaster, we encountered a number of situations in which the vast flow of unauthorized information was problematics. To assess the credibility and importance of a piece of information, we find that it is important to analyze the statement structure on Twitter and to understand the background of information. In this study, we propose a method for analyzing the statement relation between a tweet and its reply or quoted tweet. More specifically, we assume that a reply or quoted tweet expresses a statement relation (e.g., agreement, rebuttal, question, other) toward the target tweet, and we build a classifier for predicting a statement relation for a given pair of tweets. The experimental results report the performance of the classifier for predicting statement relations. In addition, we demonstrate that the proposed method can be applied to analyze statement relations between tweets that have no direct reply/quoting link, and we compare the proposed approach with the previous method based on textual entailment.

View full abstract

During the 2011 East Japan Earthquake and Tsunami Disaster, a considerable amount of false information was disseminated on Twitter; for example, after the Cosmo Oil fire, it was rumored that harmful substances will come down with rain. This paper exhaustively extracts pieces of false information from tweets within one week after the earthquake, and analyzes the diffusion of false information and its correction information. By designing a set of linguistic patterns that correct false information, this paper proposes a method for detecting false information. Specifically, the method extracts text passages that match the correction patterns, clusters the passages into topics of false information, and selects, for each topic, a passage explaining the false information most suitably. We report the performance of the proposed method on the data set extracted manually from websites that specialize in collecting false information. In addition, we build a system that visualizes emergences, diffusions, and terminations of a piece of false information and its correction. We also propose a method for discriminating false information from its correction, and discuss the possibility of alerting against false information.

View full abstract

Microblogs have enabled us to exchange information in real time, which has led to the spread of not only beneficial but also potentially harmful information, such as rumors. Rumors may block the process of adequate information sharing, which may, in turn, cause serious problems. Several studies have already analyzed the impact of rumors on microblogging media; however, the way in which these rumors can cause potential problems largely remains unclear. This paper analyzes people’s perceptions of rumors on Twitter during disasters using subjective evaluation and rhetorical unit analysis. The results showed that many subjects perceived rumors as containing information that instigates people to take action, reports on a current situation, or predicts future events. Moreover, information that instigates people to take action has been perceived as beneficial in some contexts, while it is also seen as harmful in other cases.

View full abstract

Disasters may cause a variety of health problems in the victim population, and public health authorities are forced to assess such situations rapidly in order to take appropriate countermeasures. This process may involve the processing of numerous unstructured texts, and hence, natural language processing (NLP) has significant application potential in the field of crisis response. This report classifies the information related to public health in a crisis situation into three categories—victims, victim groups, and care providers—and summarizes the characteristics of these categories to clarify the tasks suitable for NLP. This analysis is followed by three case studies of the Great East Japan Earthquake response. These case studies illustrate the contribution of NLP in an actual health crisis and suggest that the authorities do not possess appropriate means to process the texts that may accumulate in such a situation. The archive of the earthquake would be the best source for the analysis to prepare for future disasters.

View full abstract