Policy-driven Knowledge Selection and Response Generation for Document-grounded Dialogue

TL;DR

A novel framework exploiting a dialogue policy for two core tasks in DGD, namely, knowledge selection and response generation, consists of two modules: the policy planner leverages policy-aware dialogue representation to select knowledge and predict the policy of the response; the generator uses policy/knowledge-aware dialogue representation for response generation.

Abstract

Document-grounded dialogue (DGD) uses documents as external knowledge for dialogue generation. Correctly understanding the dialogue context is crucial for selecting knowledge from the document and generating proper responses. In this paper, we propose using a dialogue policy to help the dialogue understanding in DGD. Our dialogue policy consists of two kinds of guiding signals: utterance function and topic transfer intent. The utterance function reflects the purpose and style of an utterance, and the topic transfer intent reflects the topic and content of an utterance. We propose a novel framework exploiting our dialogue policy for two core tasks in DGD, namely knowledge selection (KS) and response generation (RG). The framework consists of two modules: the Policy planner leverages policy-aware dialogue representation to select knowledge and predict the policy of the response; the generator uses policy/knowledge-aware dialogue representation for response generation. Our policy-driven model gets state-of-the-art performance on three public benchmarks and we provide a detailed analysis of the experimental results. Our code/data will be released on GitHub.

Figures (9)

• Figure 1: A document-grounded dialogue example in the Wizard-of-Wikipedia dataset. For each agent turn, up to the last three turns are used as the dialogue context. There is a group of external knowledge for each agent turn, which contains multiple knowledge sentences under multiple topics. The bold sentence in each external knowledge group means it is selected for constructing the agent turn.
• Figure 2: The architecture of the PD-DGD model. Notice that $C_i$ and $K_i$ are sentences with multiple tokens.
• Figure 3: Comparison of the DA prediction accuracy and the KS improvement ratio.
• Figure 4: Comparison of the Topic transfer intent prediction accuracy and the KS improvement ratio.
• Figure 5: Comparison of the DA prediction accuracy and the RG improvement ratio.