Comparison of Vocabulary Features Among Multiple Data Sources for Constructing a Knowledge Base on Disaster Information

This research aims to develop a framework for smoothly obtaining disaster information from multiple web services through a knowledge base of disaster information. In Japan, where natural disasters occur frequently, there is a need for a system that can utilize disaster information transmitted on the Web from various locations in disaster-stricken areas for rescue operations and disaster recovery when a disaster occurs. Since such information is posted to many web services, searchers must refer to multiple web services to obtain the desired infor mation. In this study, we propose understanding the characteristics of disaster information posted on each web service and using them as a guide for searchers to obtain disaster information smoothly. To achieve this goal, we tried to construct a vocabulary set of disaster information by acquiring textual information from two different data sources and using word embedding and clustering. Comparison of the acquired disaster information revealed two points: The composition of dis aster information categories differs among data sources. Even texts in the same category have different characteristics of words depending on the data source.