Abstract:<p>The end-to-end open-domain dialogue system is a challenging task since the existing neural models suffer from the issue of trivial responses. Employing background knowledge as a major solution, has been proven to be effective to improve the responses quality. However, less attention was paid to the selection of the appropriate knowledge in scenarios where the utterance subject drifts between two partners, which could prohibit the model from learning to access knowledge correctly. In this paper, we propose a novel Knowledge Augmented Dialogue Generation (KADG) model to facilitate both knowledge selection and incorporation in open-domain dialogue systems. The core components of KADG consist of Divergent Knowledge Selector (DKS) and Knowledge Aware Decoder (KAD). DKS performs a one-hop subject reasoning over knowledge by pre-optimizing each knowledge candidate with inferred drift clue. Drift clue implies the potential subjects association of the current conversation and is served to bridge the subject gap in the knowledge selection. Thereafter, KAD makes full use of this selected knowledge to generate responses contextual coherently as well as knowledgeably. Comprehensive experiments on a newly released knowledge-grounded conversation dataset Wizard-of-Wikipedia have verified the superiority of our model than previous baselines and shown that our method can refer to the knowledge properly and generate diverse and informative responses.</p>

Knowledge dialogue generation with multi-head attention mechanism

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

Knowledge-aware Dialogue Generation with Hybrid Attention (Student Abstract)

Survey of Research on Knowledge-Driven Dialogue Generation Models

Difference-aware Knowledge Selection for Knowledge-grounded Conversation Generation

Read Then Respond: Multi-granularity Grounding Prediction for Knowledge-Grounded Dialogue Generation.

Improving knowledge-based dialogue generation through two-stage knowledge selection and knowledge selection-guided pointer network

Get The Point of My Utterance! Learning Towards Effective Responses with Multi-Head Attention Mechanism

Improving Knowledge-aware Dialogue Generation Via Knowledge Base Question Answering

Knowledge-Grounded Dialogue with Reward-Driven Knowledge Selection

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

KSAM: Infusing Multi-Source Knowledge into Dialogue Generation via Knowledge Source Aware Multi-Head Decoding

HKA: A Hierarchical Knowledge Attention Mechanism for Multi-Turn Dialogue System

Learning to Select Knowledge for Response Generation in Dialog Systems

Topic-Aware Dialogue Generation with Two-Hop Based Graph Attention.

Knowledge Grounded Pre-Trained Model for Dialogue Response Generation.

More is Better: Enhancing Open-Domain Dialogue Generation via Multi-Source Heterogeneous Knowledge

Generative Knowledge Selection for Knowledge-Grounded Dialogues

Knowledge Augmented Dialogue Generation with Divergent Facts Selection

Knowledge-aware Attentive Wasserstein Adversarial Dialogue Response Generation

Select, Extend, and Generate: Generative Knowledge Selection for Open-Domain Dialogue Response Generation.