Knowledge Dialogue Generation paper reading

学习记录。

Pretraining

Plato: Pre-trained dialogue generation model with discrete latent variable. Bao, Siqi, Huang He, Fan Wang, Hua Wu, and Haifeng Wang. (ACL 2020) paper

文章提出了一个对话生成的预训练模型，核心在于估计response act的隐变量，并采样隐变量作为transformer输入序列的special token表示。（没猜错的话，NLL loss和BOW loss均为隐变量的条件期望）

Commonsense Knowledge

Commonsense Knowledge Aware Conversation Generation with Graph Attention. Zhou, Hao, Tom Young, Minlie Huang, Haizhou Zhao, Jingfang Xu, and Xiaoyan Zhu. (IJCAI 2018) paper

文章首次将大规模commonsense knowledge融合到对话生成中，首次将knowledge以KG形式，attention方式融合到对话生成中。（基于KG的对话生成经典文章）

Diverse and informative dialogue generation with context-specific commonsense knowledge awareness. Wu, Sixing, Ying Li, Dawei Zhang, Yang Zhou, and Zhonghai Wu. (ACL 2020) paper

文章用dialogue context增强了KG的表示，并且用context和response分别构建了knowledge的先验和后验分布，用KL散度作为loss使两者相似。

Grounded Conversation Generation as Guided Traverses in Commonsense Knowledge Graphs. Zhang, Houyu, Zhenghao Liu, Chenyan Xiong, and Zhiyuan Liu. (ACL 2020) paper

文章考虑2-hop的KG关联（过去通常都是与当前turn直接关联的concept），增强模型实现concept shift的能力。

Improving Knowledge-Aware Dialogue Generation via Knowledge Base Question Answering. Wang, Jian, Junhao Liu, Wei Bi, Xiaojiang Liu, Kejing He, Ruifeng Xu, and Min Yang. (AAAI 2020) paper

文章用KBQA任务预训练对话生成模型，从而提升knowledge selection的能力。decoder端融合knowledge selection表示、encoder表示以及基于相似问题top-retrieved回复的表示（response guiding attention，一定程度解决one-to-many）。

Knowledge Document

Knowledge-Grounded Dialogue Generation with Pre-trained Language Models. Zhao, Xueliang, Wei Wu, Can Xu, Chongyang Tao, Dongyan Zhao, and Rui Yan. (EMNLP 2020) paper

文章构建了一种融合knowledge document的对话生成模型，其中document选择和对话生成共同进行优化，用RL方法无监督地调整document选择的模型。（无需对相关knowledge document进行人工标记）

Bridging the Gap between Prior and Posterior Knowledge Selection for Knowledge-Grounded Dialogue Generation. Chen, Xiuyi, Fandong Meng, Peng Li, Feilong Chen, Shuang Xu, Bo Xu, and Jie Zhou. (EMNLP 2020) paper

文章在训练阶段用后验信息增强先验信息，并设计蒸馏模型以消除exposure bias。

Sequential Latent Knowledge Selection for Knowledge-Grounded Dialogue. Kim, Byeongchang, Jaewoo Ahn, and Gunhee Kim. (ICLR 2020) paper

文章同样在训练阶段用后验信息增强先验信息，此外将knowledge selection作为序列决策过程，构建多轮knowledge selection的联合推断。

Low-Resource Knowledge-Grounded Dialogue Generation. Zhao, Xueliang, Wei Wu, Chongyang Tao, Can Xu, Dongyan Zhao, and Rui Yan. (ICLR 2020) paper

文章探索了low-resource下的对话生成问题，主要贡献是pretraining和disentangle的decoding学习。其中pretraining训练95%的参数，fine tune时固定上述参数，只调整剩下5%的参数。另外decoding manager综合考虑context、knowledge和general language model进行最终单词生成。

Learning to Select Knowledge for Response Generation in Dialog Systems. Lian, Rongzhong, Min Xie, Fan Wang, Jinhua Peng, and Hua Wu. (IJCAI 2021) paper

文章用knowledge的先验和后验分布优化knowledge selection过程。

Unclassified

Knowledge-Aware Dialogue Generation via Hierarchical Infobox Accessing and Infobox-Dialogue Interaction Graph Network. Wu, Sixing, Minghui Wang, Dawei Zhang, Yang Zhou, Ying Li, and Zhonghai Wu. (IJCAI 2021) paper

文章将infobox中的信息作为knowledge，并构建GAT网络。

Conversational Graph Grounded Policy Learning for Open-Domain Conversation Generation. Xu, Jun, Haifeng Wang, Zheng-Yu Niu, Hua Wu, Wanxiang Che, and Ting Liu. (ACL 2020) paper

Response Generation

Graph-Structured Context Understanding for Knowledge-grounded Response Generation. Li, Yanran, Wenjie Li, and Zhitao Wang. (SIGIR 2021 short) paper

文章构建了context graph，涉及utterance节点（u）和entity mention节点（m），节点通过GCN进行融合交互学习，response generation参照过去研究，从候选entity中进行copy。

Response-anticipated memory for on-demand knowledge integration in response generation. Tian, Zhiliang, Wei Bi, Dongkyu Lee, Lanqing Xue, Yiping Song, Xiaojiang Liu, and Nevin L. Zhang. (ACL 2020) paper