Abstract:Knowledge-grounded dialogue (KGD) learns to generate an informative response based on a given dialogue context and external knowledge (\emph{e.g.}, knowledge graphs; KGs). Recently, the emergence of large language models (LLMs) and pre-training techniques has brought great success to knowledge-grounded dialogue. However, when building KGD systems in real applications, there are various real-world noises that are inevitable to face. For example, the dialogue context might involve perturbations such as misspellings and abbreviations. In addition, KGs typically suffer from incompletion and also might contain erroneous and outdated facts. Such real-world noises pose a challenge to the robustness of KGD systems and hinder their applications in the real world. In this paper, we propose an entity-based contrastive learning framework for improving the robustness of KGD. Specifically, we make use of the entity information in a KGD sample to create both its positive and negative samples which involve semantic-irrelevant and semantic-relevant perturbations, respectively. The contrastive learning framework ensures the KGD model is aware of these two types of perturbations, thus generating informative responses with the potentially noisy inputs in real applications. Experimental results on three benchmark datasets show that our method achieves new state-of-the-art performance in terms of automatic evaluation scores, verifying its effectiveness and potentiality. Furthermore, we show that our method can generate better responses than comparison models in both the noisy and the few-shot settings.

What problem does this paper attempt to address?

The paper aims to address the robustness issues faced by Knowledge-Grounded Dialogue (KGD) systems in practical applications. Specifically, although existing KGD systems have made significant progress through large-scale language models (LLMs) and pre-training techniques, they still have shortcomings when dealing with real-world noise. These noises include spelling errors, abbreviations in the dialogue context, and incomplete information, incorrect facts, or outdated information in knowledge graphs (KGs). These issues pose challenges to the robustness of KGD systems, limiting their application in real-world scenarios. To address the above problems, the authors propose an entity-based contrastive learning framework (EnCo) to enhance the robustness of KGD systems. This framework adds semantically irrelevant and semantically relevant perturbations to the given dialogue samples and their related knowledge, enabling the KGD model to generate meaningful responses when faced with potentially noisy inputs. Experimental results show that this method achieves the latest state-of-the-art performance on three benchmark datasets and outperforms existing methods in terms of automatic evaluation metrics, validating its effectiveness and potential. Additionally, the method can generate better responses in noisy environments and few-shot settings.

Improving the Robustness of Knowledge-Grounded Dialogue via Contrastive Learning

Enhancing Text-based Knowledge Graph Completion with Zero-Shot Large Language Models: A Focus on Semantic Enhancement

Multi-level Adaptive Contrastive Learning for Knowledge Internalization in Dialogue Generation

RT-KGD: Relation Transition Aware Knowledge-Grounded Dialogue Generation

Small Changes Make Big Differences: Improving Multi-turn Response Selection in Dialogue Systems Via Fine-Grained Contrastive Learning

A Contrastive Framework for Enhancing Knowledge Graph Question Answering: Alleviating Exposure Bias

GLM-Dialog: Noise-tolerant Pre-training for Knowledge-grounded Dialogue Generation

Knowledge Augmented Dialogue Generation with Divergent Facts Selection

Exploring In-Context Learning for Knowledge Grounded Dialog Generation

Contextual Knowledge Learning For Dialogue Generation

KG-ECO: Knowledge Graph Enhanced Entity Correction for Query Rewriting

Infusing internalized knowledge of language models into hybrid prompts for knowledgeable dialogue generation

GLSEC:Global and local semantic-enhanced contrastive framework for knowledge graph completion

Entity-Consistent End-to-end Task-Oriented Dialogue System with KB Retriever

SimKGC: Simple Contrastive Knowledge Graph Completion with Pre-trained Language Models

Improving Knowledge Graph Completion with Structure-Aware Supervised Contrastive Learning

Knowledge Graph Contrastive Learning Based on Relation-Symmetrical Structure

KGPA: Robustness Evaluation for Large Language Models via Cross-Domain Knowledge Graphs

Robustness Testing of Language Understanding in Task-Oriented Dialog

PluDG: enhancing task-oriented dialogue system with knowledge graph plug-in module

Mining Implicit Entity Preference from User-Item Interaction Data for Knowledge Graph Completion Via Adversarial Learning