Design Representation as Semantic Networks

TL;DR

This paper introduces a method to automatically generate semantic network representations of design concepts using a large-scale pre-trained knowledge base, improving scalability and reducing manual effort in design representation.

Contribution

It presents a novel methodology that leverages pre-trained knowledge bases to create semantic network representations from natural language design descriptions, eliminating the need for ad hoc statistics.

Findings

Semantic networks effectively capture design information.

Automatic generation shows comparable quality to manual methods.

Study reveals differences based on knowledge base choices.

Abstract

Design representation is a common task in the design process to facilitate learning, analysis, redesign, communication, and other design activities. Traditional representation techniques rely on human expertise and manual construction and are difficult to repeat and scale. Here, we propose a methodology that utilizes a pre-trained large-scale cross-domain design knowledge base to automatically generate design representation as a semantic network, i.e., a network of the entities and relations, based on design descriptions in texts or natural languages. Our methodology requires no ad hoc statistics. Based on a participatory study, we reveal the effectiveness and differences of the semantic network representations that are automatically generated with alternative knowledge bases. The findings illuminate future research directions to enhance design representation as semantic networks.