An enhanced semantic expansion fabric retrieval system based on knowledge graphs

Current fabric retrieval systems face the critical issue of transforming user input into domain-specific knowledge for retrieval. Traditional text-based retrieval methods rely on the predefined and annotated fabric attribute values, which are different from textual descriptions formed by users’ perception of fabric. Establishing a connection between user descriptions and professional terms to precisely grasp user demands is the key to achieving precise text-based retrieval. To address this challenge, this paper proposes an enhanced semantic expansion fabric retrieval system based on knowledge graph. First, the sequence labeling algorithm extracts knowledge from unstructured fabric data, which is then integrated with structured datasets to build a fabric knowledge graph. Next, we design a knowledge graph matching method based on multiscale neighbor sampling (MSNS) to extend the keyword matching retrieval. Finally, a label learning derivative network (LLDN) is employed to learn labels and assign them to previously unlabeled images. An experimental evaluation on a real fabric dataset from a textile company shows that the area under the receiver operating characteristic curve of click-through rate (CTR) prediction improves by 6.7% and the accuracy improves by 3.3% compared with the baseline model. Top-K retrieval improves Precison@K by more than 5% and Recall@K by more than 10%. This substantiates the feasibility of the approach and assists enterprises in more effectively managing and leveraging fabric data.