PK-Chat: Pointer Network Guided Knowledge Driven Generative Dialogue Model
Cheng Deng, Bo Tong, Luoyi Fu, Jiaxin Ding, Dexing Cao, Xinbing Wang, Chenghu Zhou
TL;DR
Open-domain dialogue systems often struggle to provide domain-specific, accurate knowledge, suffering from out-of-vocabulary and hallucination issues when relying solely on parametric knowledge. The authors introduce PK-Chat, which fuses a unified pretrained language model with a pointer-network-based generation mechanism and a knowledge-graph retriever to generate text that can cite specific graph facts. Key components include a discrete hidden variable $z \in \{1,\dots,K\}$ governing semantic behavior and a generation probability $P(r|c,k,z)$, combined with a pointer mechanism yielding $P(w) = \lambda_{\mathrm{gen}} P_{\mathrm{vocab}}(w) + (1-\lambda_{\mathrm{gen}}) \sum_{i:w_i=w} a_i^t$. The GA-Dialogue benchmark built on GAKG (via Neo4J) and augmented with DuConv and DuRecDial enables evaluation of academic dialogue agents, and empirical results on GA-Dialogue, Persona-Chat, and DailyDialog show improvements in automatic metrics (e.g., BLEU, Distinct) and in human judgments.
Abstract
In the research of end-to-end dialogue systems, using real-world knowledge to generate natural, fluent, and human-like utterances with correct answers is crucial. However, domain-specific conversational dialogue systems may be incoherent and introduce erroneous external information to answer questions due to the out-of-vocabulary issue or the wrong knowledge from the parameters of the neural network. In this work, we propose PK-Chat, a Pointer network guided Knowledge-driven generative dialogue model, incorporating a unified pretrained language model and a pointer network over knowledge graphs. The words generated by PK-Chat in the dialogue are derived from the prediction of word lists and the direct prediction of the external knowledge graph knowledge. Moreover, based on the PK-Chat, a dialogue system is built for academic scenarios in the case of geosciences. Finally, an academic dialogue benchmark is constructed to evaluate the quality of dialogue systems in academic scenarios and the source code is available online.