Kansei engineering for the functional design of new energy vehicle: a graph-theory approach with LLM and online data

Kansei engineering (KE) links users’ perceptional requirements with the parametric design configuration, and it has been widely used in the automobile industry. This paper intents to explore the possibility to apply the AI-assisted large language model (LLM) to elicit useful KE knowledge from the internet data, such as the user-generated-content (UGC) from the automobile online communities. In particular, we focus on the new energy vehicle (NEV) which has been a trend in the Chinese market. First, online UGCs are crawled from the DongCheDi website. Then, a DeepSeek-R1 is deployed locally to process the UGC, so as to find the “mentioned function / opinion” pairs, which was not satisfyingly done by BERT models. Then, the SnowNLP model is fine-tuned and applied to quantify the emotional score in the opinion, and a word2vec model is adopted to group similar functions into categories. A configuration network is developed, where nodes are the NEV functions, and the in-arc of the node is the averaging emotional score of the function. The graph theory is employed: (1) The shortest-path algorithm is used to find the most-satisfying NEV design configuration. (2) The in-degree of node is used to quantify the importance of the NEV function. (3) The community detection algorithm is used to find the functional clusters that were mostly mentioned simultaneously. Our method can provide insights for designers, engineers, and marketing analysts on the new product development of NEV.