Grounded Conversation Generation as Guided Traverses in Commonsense Knowledge Graphs

Houyu Zhang, Zhenghao Liu, Chenyan Xiong, Zhiyuan Liu
11/7/2019
cs.CLcs.AI

Official code: thunlp/ConceptFlow (120 stars, MIT, last commit 2023-02-17).

Abstract

Human conversations naturally evolve around related concepts and scatter to multi-hop concepts. This paper presents a new conversation generation model, ConceptFlow, which leverages commonsense knowledge graphs to explicitly model conversation flows. By grounding conversations to the concept space, ConceptFlow represents the potential conversation flow as traverses in the concept space along commonsense relations. The traverse is guided by graph attentions in the concept graph, moving towards more meaningful directions in the concept space, in order to generate more semantic and informative responses. Experiments on Reddit conversations demonstrate ConceptFlow's effectiveness over previous knowledge-aware conversation models and GPT-2 based models while using 70% fewer parameters, confirming the advantage of explicit modeling conversation structures. All source codes of this work are available at https://github.com/thunlp/ConceptFlow.

View on arXivView PDF

Code Implementations(1)

thunlp/ConceptFlowOfficial100%
MIT
12018PythonApr 29, 20203 years agoMIT

Similar Papers

Proactive Human-Machine Conversation with Explicit Conversation Goals

88% match1 repo6.9k2019

C5: Towards Better Conversation Comprehension and Contextual Continuity for ChatGPT

85% match2023

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

84% match1 repo82022

Bridging Information Gaps in Dialogues With Grounded Exchanges Using Knowledge Graphs

83% match1 repo42024

Incremental Transformer with Deliberation Decoder for Document Grounded Conversations

82% match1 repo862019

Knowledge-Grounded Conversational Data Augmentation with Generative Conversational Networks

81% match2022

Cite this paper

@article{zhang2019grounded,
  title  = {Grounded Conversation Generation as Guided Traverses in Commonsense Knowledge Graphs},
  author = {Houyu Zhang and Zhenghao Liu and Chenyan Xiong and Zhiyuan Liu},
  year   = {2019},
  eprint = {1911.02707},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1911.02707}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1911.02707.svg)](https://codeofpaper.com/paper/1911.02707)

Discussion