A RAG-based framework for material matching in LCA: integrating semantic and character similarity with AI-driven explanations

In early building planning, the selection of appropriate materials is crucial to reducing emissions. To solve the problem of manually searching for suitable materials in the Life Cycle Assessment (LCA) database, some studies have developed automatic matching methods. These methods are based on Deep Learning (DL)-based Natural Language Processing (NLP). However, the outputs typically provide only matching results without explanations, which limits their usability for non-experts.

This study proposes a Retrieval Augmented Generation (RAG)-based material matching framework that incorporates a hybrid scoring of semantic similarity and edit distance to search similar material and employs GPT-2 to generate natural language explanations. This approach not only effectively retrieves matching materials from the LCA database but also improves the comprehensibility of the material matching results through the explanations generated by GPT-2. The RAG-based framework offers more instructive material recommendations for the case construction project and effectively supports non-construction material experts and non-LCA experts in decision-making.