An industrial robotic arm retrieval-augmented generation method (MTRAG) based on question rewriting and temporal recency

To address limitations in current industrial knowledge question-answering systems particularly in applications related to robotic arms such as insufficient utilization of question rewriting and neglecting vector temporal recency in retrieval ranking, we propose an improved retrieval-augmented generation method called MTRAG (Multi-Question and Temporal Retrieval Augmented Generation). MTRAG introduces question rewriting to generate multiple related question forms, enhancing retrieval coverage and accuracy by capturing different expressions and potential meanings. Additionally, it employs a recency-based retrieval strategy to weight retrieval results, assigning higher importance to recently accessed vectors to improve the quality and timeliness of the retrieval. Experimental results demonstrate that MTRAG significantly outperforms traditional RAG methods in RAGAS metrics, proving its effectiveness in enhancing retrieval and generation performance. This research offers a new solution for knowledge question-answering systems in the industrial robotic arm domain, with important theoretical significance and practical application value.