Improving LLM's Attachment to External Knowledge In Dialogue Generation Tasks Through Entity Anonymization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sheikhi, Hadi, Huang, Chenyang, Zaïane, Osmar R.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911266681913344
author Sheikhi, Hadi
Huang, Chenyang
Zaïane, Osmar R.
author_facet Sheikhi, Hadi
Huang, Chenyang
Zaïane, Osmar R.
contents Knowledge graph-based dialogue generation (KG-DG) is a challenging task requiring models to effectively incorporate external knowledge into conversational responses. While large language models (LLMs) have achieved impressive results across various NLP tasks, their ability to utilize external knowledge in KG-DG remains under-explored. We observe that LLMs often rely on internal knowledge, leading to detachment from provided knowledge graphs, even when they are given a flawlessly retrieved knowledge graph. First, we introduce LLM-KAT, an evaluation procedure for measuring knowledge attachment in generated responses. Second, we propose a simple yet effective entity anonymization technique to encourage LLMs to better leverage external knowledge. Experiments on the OpenDialKG dataset demonstrate that our approach improves LLMs' attachment on external knowledge.
format Preprint
id arxiv_https___arxiv_org_abs_2511_11946
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improving LLM's Attachment to External Knowledge In Dialogue Generation Tasks Through Entity Anonymization
Sheikhi, Hadi
Huang, Chenyang
Zaïane, Osmar R.
Computation and Language
Machine Learning
Knowledge graph-based dialogue generation (KG-DG) is a challenging task requiring models to effectively incorporate external knowledge into conversational responses. While large language models (LLMs) have achieved impressive results across various NLP tasks, their ability to utilize external knowledge in KG-DG remains under-explored. We observe that LLMs often rely on internal knowledge, leading to detachment from provided knowledge graphs, even when they are given a flawlessly retrieved knowledge graph. First, we introduce LLM-KAT, an evaluation procedure for measuring knowledge attachment in generated responses. Second, we propose a simple yet effective entity anonymization technique to encourage LLMs to better leverage external knowledge. Experiments on the OpenDialKG dataset demonstrate that our approach improves LLMs' attachment on external knowledge.
title Improving LLM's Attachment to External Knowledge In Dialogue Generation Tasks Through Entity Anonymization
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2511.11946