Highlights

  • home_icon HOME
  • Research
  • Highlights
Development of Big Data-Based Self-Evolving Knowledge Base and Reasoning Platform


People today search the internet to acquire knowledge. The internet is a huge ocean of information, including knowledge data of various fields, so much so that there are even competitions for internet information search in which participants compete to solve problems by collecting various data from the internet. If AI can collect data about various information included in input text that a user enters to search the internet, the desired data may be easily obtained even if the user is not highly skillful in search. The present research project was conducted to develop the first big data-based self-evolving knowledge base and reasoning platform in the Korean language.


Korean Big Data-Based Self-Evolving Knowledge Base

Kbox (Knowledge Box) is the result of symbolic-deep learning architecture-based self-evolving knowledge technology based on big data analysis and AI. As texts are entered, relevant pieces of knowledge are gathered to increase the KBox knowledge base. The word strings included in a text are connected with knowledge instances, and relations between knowledge instances are extracted (relation extraction: verified by TTA V&V) through co-reference resolution in which instances having different word strings are verified as identical instances or different ones. While optimal learning data are established through crowd sourcing, the AI system performs self-learning, understands natural language, and automatically extends the scope of self-learning and the knowledge base used as the basis of verification. When applied to conventional internet search, relevant web pages, which used to be separately searched from various websites, are collected and shown to user to increase user convenience.

The Entity Summarization generates a summary of the data collected by the internet search by combining the data with natural language generation technology. The AI extracts from the searched instances the parts that are considered as important and summarizes the extracted information. Daily information changes had to be updated manually in the past but the present technology may be applied to let the AI automatically acquire knowledge produced from various sources and update them. The present research project is significant because the self-evolving knowledge base and reasoning technology were developed for the first time for the Korean language.


Contribution to International Competitiveness of Korean Companies

The research team has expanded its expertise in the new area and developed the intelligent consultation platform technology to promote domain-specialized commercialization through institutions demanding it, securing an in-depth Q&A service of high quality that can be readily commercialized. Compared to leading countries such as the US, UK and China, and global companies including Google and IBM, investment in AI technologies has been very small in Korea. If large-scale knowledge data and the linguistic intelligence technologies developed in the present research project are commercialized rapidly, Korean companies will be helped to establish AI-based global service platforms and secure international competitiveness. In addition, the results of the present research project may be used to prevent encroachment of overseas global companies in the Korean market as well as to prevent increases of common costs in the industry by monopoly and collapse of Korean technological ecosystem due to dependence on foreign technologies.


Prof. Choi, Key-Sun
2019 KI Annual Report


KAIST 291 Daehak-ro, Yuseong-gu, Daejeon (34141)
T : +82-42-350-2381~2384
F : +82-42-350-2080
Copyright (C) 2015. KAIST Institute