ALEX:A Light Editing-knowledge Extractor

TL;DR

The paper tackles the problem of keeping large language models up-to-date by enabling scalable knowledge editing, particularly for multi-hop questions where retrieval efficiency is critical. It introduces ALEX, a lightweight framework with a hierarchical memory architecture (Semantic Manifold Partitioning), an Inferential Query Synthesis module, and a Dynamic Evidence Adjudication engine to enable fast, accurate retrieval of relevant edits. ALEX reduces retrieval complexity from $O(N)$ to $O(K+N/C)$, achieves state-of-the-art MultiHop-ACC and HopWise-ACC on MQuAKE benchmarks, and reduces the edit search space by over 80%. The results demonstrate robustness across model sizes and show strong gains on harder, deeper reasoning tasks, highlighting ALEX's potential for scalable, real-world knowledge editing in LLMs.

Abstract

The static nature of knowledge within Large Language Models (LLMs) makes it difficult for them to adapt to evolving information, rendering knowledge editing a critical task. However, existing methods struggle with challenges of scalability and retrieval efficiency, particularly when handling complex, multi-hop questions that require multi-step reasoning. To address these challenges, this paper introduces ALEX (A Light Editing-knowledge Extractor), a lightweight knowledge editing framework. The core innovation of ALEX is its hierarchical memory architecture, which organizes knowledge updates (edits) into semantic clusters. This design fundamentally reduces retrieval complexity from a linear O(N) to a highly scalable O(K+N/C). Furthermore, the framework integrates an Inferential Query Synthesis (IQS) module to bridge the semantic gap between queries and facts , and a Dynamic Evidence Adjudication (DEA) engine that executes an efficient two-stage retrieval process. Experiments on the MQUAKE benchmark demonstrate that ALEX significantly improves both the accuracy of multi-hop answers (MultiHop-ACC) and the reliability of reasoning paths (HopWise-ACC). It also reduces the required search space by over 80% , presenting a promising path toward building scalable, efficient, and accurate knowledge editing systems.

Figures (6)

• Figure 1: Sparse vs. Dense Retrieval: Speed vs. Accuracy.
• Figure 2: Workflow of our knowledge editing system.
• Figure 3: The comparison of the correct number of PokeMQA and PokeMQA (ALEX) under 2-Hop, 3-Hop, and 4-Hop conditions on the MQuAKE-CF-3K (left) and MQuAKE-CF-3K-v2 (right) datasets.
• Figure 4: Performance of the clustering and retrieval components across five datasets. The number of clusters (K=12) was optimized for these datasets as detailed in Appendix B.
• Figure 5: The figure illustrates Cluster ACC and Retrieval ACC under varying numbers of cluster classifications.