HTML5 Webook

33/64

NICT REPORT 31testbed that collects atmospheric environ-ment data (PM2.5 levels, temperature and humidity, etc.) and health data (heart rate, autonomic nervous balance, etc.) and pre-pares digital maps including charts and comments showing how they are associ-ated. A field demonstration trial with the participation of local residents aimed at finding healthy air was held in Fukuoka City between March 10 and April 8, 2018. A total of 69 people took part in this trial, in-cluding members of a running club, a re-gional open innovation group (One Japan in Kyushu), local students, and an IT volun-teer group (Code for Fukuoka). The results of a questionnaire survey showed that most of them found the study very inter-esting, and included positive comments such as “It was a good opportunity to con-sider how my own vital data is affected by the environment,” “It was interesting to see data about what I feel every day,” and “It was fun using the latest sensor technology and learning about NICT’s work.”R&D of social big data analysis infrastructureIn the R&D of large-scale information inte-grated visualization technology, we are work-ing on a 3D visualization method that facili-tates understanding of the effects of day-to-day events in the real world, and how these effects spread across space and time. By focusing on position-related expressions in micro-blogging streams (of which Twitter is a prime example), our system associates tweets with location coordinates (either di-rectly attached to the tweet or based on words such as the names of places or estab-lishments). It then produces a multi-layered 3D visualization by checking for local events and wide-area events, using animation to show how this information changes over time. We are currently working on applying this system to the analysis and visualization of heavy rain risk by integrating it with rainfall data obtained by a weather radar (Fig.3).Also, in the research and development of data mining technology targeting non-textual data, we have developed a new algorithm that recursively searches tree-structured data converted from a time-series database based on periodic-frequency indicators that mea-sure the temporal periodicity of item sets, and efficiently discovers all the item sets with par-tial periodicity. On the other hand, in large-scale graph data analysis technology, we have confirmed that the cost of communica-tion can be reduced by 12% on average by running GraphSlice (an efficient distributed processing framework for social graphs) on Apache Spark in order to develop a scalable distributed graph database engine suitable for cloud environments.Furthermore, in the R&D of human behav-ior analysis using social media data, we have succeeded in detecting tweets that strongly influence purchasing behavior with high ac-curacy (F-value of 0.53) based on multiple clues such as the relevance of the tweet con-tents, the proximity of tweet posting times, and the degree of closeness between users.Fig.3 : Integrated visualization of heavy rain data (PANDA, XRAIN) and geospatial word-clouds based on social big data Fig.2 : Experimental demonstration of smart IoT for an environment and health monitoring system aimed at improving air qualityA scene at the Datathon held in Fukuoka in March to April 2018Research and Development