@inproceedings{fu-etal-2025-knowledge,
    title = "Knowledge Graph-Driven Memory Editing with Directional Interventions",
    author = "Fu, Jinhu  and
      Wang, Kun  and
      Guo, Chongye  and
      Fang, Junfeng  and
      Zhang, Wentao  and
      Su, Sen",
    editor = "Christodoulopoulos, Christos  and
      Chakraborty, Tanmoy  and
      Rose, Carolyn  and
      Peng, Violet",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
    month = nov,
    year = "2025",
    address = "Suzhou, China",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-nlpsi/2025.findings-emnlp.261/",
    doi = "10.18653/v1/2025.findings-emnlp.261",
    pages = "4860--4874",
    ISBN = "979-8-89176-335-7",
    abstract = "Large Language Models (LLMs) have revolutionized language processing and understanding, yet their performance is hampered by inaccuracies and outdated information. Model editing techniques offer a solution but face two key challenges: \textbf{(I)} Most methods inject knowledge by constructing rigid loss, which leads to poor compatibility when dealing with higher-order multi-hop problems. \textbf{(II)} Locate-then-edit vein, by altering pre-trained parameters, inevitably affect normal knowledge and even face the catastrophic forgetting. In this paper, we introduce \textbf{KGMET}, a framework that constructs knowledge graphs using available information to guide the direction of knowledge editing, enabling \textbf{consistent}, \textbf{aligned}, and \textbf{stable} information during \textbf{large-scale} editing scenario. Furthermore, \textit{KGMET} goes beyond this by employing orthogonal constraints to block the interference of irrelevant information, ensuring the updates are both controllable and generalizable. Experiments on Multi-Conterfact, ZsRE, and MQuAKE datasets using \textit{Llama-3-8B}, \textit{GPT-J-6B}, and \textit{GPT-2-XL} models showcase improvements over state-of-the-art methods, with $\uparrow 5\%-17\%$ in multi-hop tasks while remaining generalizable (at least $\uparrow 20\%$ in fluency). Our code is available on Github."
}