Abstract:Knowledge graph-based dialogue systems are capable of generating more informative responses and can implement sophisticated reasoning mechanisms. However, these models do not take into account the sparseness and incompleteness of knowledge graph (KG)and current dialogue models cannot be applied to dynamic KG. This paper proposes a dynamic Knowledge graph-based dialogue generation method with improved adversarial Meta-Learning (KDAD). KDAD formulates dynamic knowledge triples as a problem of adversarial attack and incorporates the objective of quickly adapting to dynamic knowledge-aware dialogue generation. We train a knowledge graph-based dialog model with improved ADML using minimal training samples. The model can initialize the parameters and adapt to previous unseen knowledge so that training can be quickly completed based on only a few knowledge triples. We show that our model significantly outperforms other baselines. We evaluate and demonstrate that our method adapts extremely fast and well to dynamic knowledge graph-based dialogue generation.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are the poor performance of existing knowledge - graph - based dialogue systems when dealing with sparse and incomplete knowledge graphs (KGs), and the inability of current dialogue models to be applied to dynamic knowledge graphs. Specifically: 1. **Sparsity and Incompleteness**: Existing knowledge graphs often have insufficient information, which affects the quality of dialogue generation. 2. **Dynamic Adaptability**: Entities or relationships in dynamic knowledge graphs change over time, while existing dialogue models lack the ability to adapt to such dynamic changes. To solve these problems, the paper proposes a dynamic knowledge - graph - based dialogue generation method (KDAD) combined with Improved Adversarial Meta - Learning (ADML). This method regards the changes of dynamic knowledge triples as an adversarial attack problem and quickly adapts to new knowledge triples with a small number of samples, thereby improving the performance of the dialogue system. ### Formula Summary - **Probability Calculation in the Knowledge Selection Module**: \[ \nu_X=\arg\max P(v|X) \] \[ P(Y|X, v)=\prod_{t = 1}^{n}P(y_t|Y_{<t}, X, v) \] - **Path Matrix Calculation**: \[ R_t=\theta(\text{softmax}(d_t)) \] \[ A_{ij}=\begin{cases} \gamma&\text{if }(h, r, t)\in K\\ 0&\text{otherwise} \end{cases} \] \[ T_t = R_t\cdot A \] \[ K_t=N(T_t)\cdot s \] - **Updated Formulas for Improved Adversarial Meta - Learning**: \[ \theta'_{\text{clean}}^i=\theta-\alpha_1\nabla_\theta\mathcal{L}_i(M_\theta, D_{\text{clean}}^i) \] \[ \theta'_{\text{adv}}^i=\theta-\alpha_2\nabla_\theta\mathcal{L}_i(M_\theta, D_{\text{adv}}^i) \] \[ \theta=\theta-\beta_1\nabla_\theta\sum_{\mathcal{T}_i\sim\mathcal{T}}\mathcal{L}_i(M_{\theta'_{\text{adv}}^i}, D_{\text{clean}}'^i) \] \[ \theta=\theta-\beta_2\nabla_\theta\sum_{\mathcal{T}_i\sim\mathcal{T}}\mathcal{L}_i(M_{\theta'_{\text{clean}}^i}, D_{\text{clean}}'^i) \] Through these formulas, the paper shows how to use the improved adversarial meta - learning algorithm to improve the performance of the dialogue generation system based on dynamic knowledge graphs.

Dynamic Knowledge Graph-based Dialogue Generation with Improved Adversarial Meta-Learning

Knowledge-graph based Proactive Dialogue Generation with Improved Meta-learning

Adversarial Learning for Neural Dialogue Generation.

Read Key Points: Dialogue-Grounded Knowledge Points Generation with Multi-Level Salience-Aware Mixture

Knowledge Augmented Dialogue Generation with Divergent Facts Selection

There Is No Standard Answer: Knowledge-Grounded Dialogue Generation with Adversarial Activated Multi-Reference Learning

DyKgChat: Benchmarking Dialogue Generation Grounding on Dynamic Knowledge Graphs

Knowledge Graph-Augmented Language Models for Knowledge-Grounded Dialogue Generation

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

KRP-DS: A Knowledge Graph-Based Dialogue System with Inference-Aided Prediction

Learning Symmetric Collaborative Dialogue Agents with Dynamic Knowledge Graph Embeddings

Knowledge-aware Attentive Wasserstein Adversarial Dialogue Response Generation

Contextual Knowledge Learning For Dialogue Generation

Enhancing Dialogue Generation via Dynamic Graph Knowledge Aggregation

Topic-level knowledge sub-graphs for multi-turn dialogue generation

DAL: Dual Adversarial Learning for Dialogue Generation.

Open-domain Dialogue Generation Grounded with Dynamic Multi-form Knowledge Fusion

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

Improving Knowledge-aware Dialogue Generation Via Knowledge Base Question Answering

Generative Knowledge Selection for Knowledge-Grounded Dialogues

Dynamically Retrieving Knowledge via Query Generation for Informative Dialogue Generation