Cost-Aware Outdated Facts Correction in the Knowledge Bases

Hao Xin, Lei Chen, Yanyan Shen

Abstract: In recent years, retrieval augmented generation model (RAG) outperform other models in the tasks that required a lot of knowledge. It helps large language models to retrieve relevant information from data sources and incorporate it into generated text. Knowledge bases (KBs) is one of the main data sources used by RAG model. However, the facts stored in KBs can also get out-of-date and therefore need to be updated from time to time. In spite of its importance, the knowledge freshness problem in KBs is relatively unexplored. Existing approaches such as directly extracting new fact from news can only cover a small portion of them. In this work, we propose a framework to discover and update the outdated facts in KBs. Specifically, we first try to identify the outdated facts by utilizing features like update frequency. In order to check and update all of the facts with minimal cost, we propose a cost-aware fact selection algorithm to guide the fact update process. We also generate explainable rules to find the dependency with the fact update to minimize the cost. We have conducted extensive experiments on real KBs. The experimental results verify the effectiveness and efficiency of our proposed framework.