Network design system recommending similar configurations characterized by manual knowledges

Recently, the number of skilled network design engineers is not enough, and the improvement of non-expert's working efficiency and design quality by the design support system becomes an important issue. Therefore, we developed the network design support system that recommended similar configuration files when the engineer made new design. Our system converts the dependent part of each design matter in existing configuration files into the representation that showed design characteristics by use of the dictionary extracted from the manual of network device. It performs clustering (unsupervised learning) of the converted configuration files, and finds similar ones. We report the concept and implementation of the developed system.