HTML5 Webook

14/64

12Data utilizaon and analytics plaormUniversal Communication Research Institute We are conducting R&D to contribute to increasing convenience for people and building a rich and secure society through human-friendly communication technology and intelligent advanced technology. In particular, we are conduct-ing R&D on technology that uses big data, incorporating the vast amount of knowledge and information circulating in society (social knowledge) as a source of information, and makes specialist knowledge available easily, even to non-specialists. This is done by generating useful questions and automatically providing answers to those questions, and by providing knowledge that helps users in making decisions. We also engage in joint R&D with the Resilient ICT Research Center, on platform technology that will organize the social knowledge on the Internet regarding disasters in real time, integrate it with various types of observation data, and to provide it to users in an easy to understand form. To optimize and increase the eiciency of various social systems, we are also conducting R&D on an image analysis technology providing advanced recognition of circumstances and support for taking action. We aim to create new ICT that will realize human-friendly and society-friendly communication and be useful for the lifestyles and well-being of people.Application and use of DISAANA/D-SUMM (Joint development with the Resilient ICT Research Center)Details regarding DISAANA and D-SUMM will be described in the section on the Resilient ICT Research Center.Next-generation conversation technology R&DTo use social knowledge effectively, we are promoting development of the WEKDA next-generation dialog system. WEKDA, or WEb-based Knowledge Disseminating dialog Agent, uses the large volume of knowledge on the Web to conduct conversation on a wide range of topics (Fig.1). Rather than con-versing based on rules and scenarios, as with other conversation technologies, it uses questions and answers based on the vast in-formation on the Web. There is demand in society for this sort of advanced next genera-tion artificial intelligence technology. Underly-ing WEKDA is WISDOM X, a system that ana-lyzes the information on some four billion Web pages to present answers to questions. When given an input phrase (e.g.: “iPS cells are amazing, aren’t they?”), it uses deep learn-search results, using a deep learning method similar to machine translation. These enabled the system to generate more appropriate re-sponses from the results of the answer search. The architecture of WEKDA is shown in Fig.2. The dialog system was implemented by combining multiple deep-learning mod-ules, including those just mentioned. Ques-tions can be input directly to WISDOM X and the answer will be presented. Currently, WEKDA is limited to handling “What” questions that can be answered with Fig.1 : Conversation with the WEKDA next-generation dialog systeming to ask a question about what to show the user (e.g.: “What would you like to see about iPS cells?). When WISDOM X is given this question, the system generates a response from the results of searching for an answer (e.g.: “A possible treatment for cardiomyopa-thy using iPS cells has been found.”). An out-line of this mechanism was developed in FY2016, and in FY2017, we implemented a deep learning mechanism to rank the results from the answer search and also functionality to generate an appropriate response from the Research and Development