一文速览知识增强的对话推荐系统
本文主要介绍了知识增强的对话推荐系统。文章也同步发布在AI Box知乎专栏（知乎搜索 AI Box专栏），欢迎大家在知乎专栏的文章下方评论留言，交流探讨！
引言：对话推荐系统（CRS）旨在通过交互式的语言对话为用户提供推荐服务，它通常包括推荐和对话两个模块。传统的对话系统大多是一个端到端系统，将历史对话作为输入来产生应答，这往往利用语料库中的高频词产生回复，这种“安全”的回答往往缺乏有意义的信息。因此，基于知识增强的对话推荐系统被提出了，系统可以基于对话语境和外部知识，生成具有丰富信息并有意义的回答。
1 相关数据集
Wizard of Wikipedia: Knowledge-Powered Conversational Agents
Towards Policy-Guided Conversational Recommendation with Dialogue Acts
Towards Conversational Recommendation over Multi-Type Dialogs（DuRecDial）
DuRecDial 2.0: A Bilingual Parallel Corpus for Conversational Recommendation
NaturalConv: A Chinese Dialogue Dataset Towards Multi-turn Topic-driven Conversation
KdConv: A Chinese Multi-domain Dialogue Dataset Towards Multi-turn Knowledge-driven Conversation
A Large-Scale Chinese Short-Text Conversation Dataset（LCCC）
Towards Expressive Communication with Internet Memes: A New Multimodal Conversation Dataset and Benchmark （MOD）
Human-to-Human Conversation Dataset for Learning Fine-grained Turn-taking Action（FTAD）
Towards Empathetic Open-domain Conversation Models: a New Benchmark and Dataset
CrossWOZ: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset
Adding Chit-Chat to Enhance Task-Oriented Dialogues（ACCENTOR）
SalesBot: Transitioning from Chit-Chat to Task-Oriented Dialogues
Recommendation as a Communication Game: Self-Supervised Bot-Play for Goal-oriented Dialogue （GoRecDial）
OpenDialKG: Explainable Conversational Reasoning with Attention-based Walks over Knowledge Graphs
INSPIRED: Toward Sociable Recommendation Dialog Systems
INSPIRED2: An Improved Dataset for Sociable Conversational Recommendation
Towards Topic-Guided Conversational Recommender System（TGRedial）
2. 不同知识来源的对话增强
2.1 通过主题进行增强
在对话系统中，为了避免产生琐碎并且偏离主题的回复，基于主题增强的对话系统可以从潜在的主题中挖掘文档的高阶语义信息。
Topic aware neural response generation
A Neural Topical Expansion Framework for Unstructured Persona-oriented Dialogue Generation
定义 | 优点 | 缺点 | |
|---|---|---|---|
概率模型 | 将文档和单词的语义表示整合到一个网络中 | 严格的概率解释 | 训练和生成过程分离，无视输入和输出的依赖 |
神经网络模型 | 结合神经网络和概率主题模型 | 更好主题连贯性，可以反向传播进行训练 | 忽略了文档之间的相关性 |
2.2 通过关键词进行增强
关键词：一个或多个单词的序列，为文档主要内容。
Sequence to Backward and Forward Sequences: A Content-Introducing Approach to Generative Short-Text Conversation
Generating responses with a specific emotion in dialog
Emotion-aware Chat Machine: Automatic Emotional Response Generation for Human-like Emotional Interaction
Emotional chatting machine: Emotional conversation generation with internal and external memory
定义 | 优点 | 缺点 | 适用范围 | |
|---|---|---|---|---|
关键词分配方法 | 从预定义的词表中，给每个输入文档分配关键词 | 1. 相似文档可能用不同关键词 | 1. 新领域创造字典很昂贵 | 需要特定类别关键词引导文本生成 |
关键词提取方法 | 从输入文档中提取一个词作为关键词 | 不受环境约束，可以广泛适用 | 1. 相似文档可能用不同关键词 | 输出序列中需要依靠输入中的重要信息 |
2.3 通过知识库进行增强
知识库存储，管理了大量大规模的信息，里面包含大量的三元组知识。
A Copy-Augmented Sequence-to-Sequence Architecture Gives Good Performance on Task-Oriented Dialogue
Mem2Seq: Effectively Incorporating Knowledge Bases into End-to-End Task-Oriented Dialog Systems
Learning to select knowledge for response generation in dialog systems
TopicKA: Generating Commonsense Knowledge-Aware Dialogue Responses Towards the Recommended Topic Fact
Improving Knowledgeaware Dialogue Generation via Knowledge Base Question Answering
Diverse and Informative Dialogue Generation with Context-Specific Commonsense Knowledge Awareness
2.4 通过知识图谱进行增强
知识图谱是一种结构化的人类知识，由许多的知识三元组构成的。三元组由实体，关系和结构化介绍组成的。
Knowledge Aware Conversation Generation with Explainable Reasoning over Augmented Graphs
DyKgChat: Benchmarking Dialogue Generation Grounding on Dynamic Knowledge Graphs
Grounded conversation generation as guided traverses in commonsense knowledge graphs
Commonsense knowledge aware conversation generation with graph attention
2.5 通过文本进行增强
Wizard of wikipedia: Knowledge-powered conversational agents
A knowledge-grounded neural conversation model
RefNet: A referenceaware network for background based conversation
Conversing by Reading: Contentful Neural Conversation with On-demand Machine Reading
Thinking Globally, Acting Locally: Distantly Supervised Global-to-Local Knowledge Selection for Background Based Conversation
DEEPCOPY: Grounded Response Generation with Hierarchical Pointer Networks
更多推荐
100篇论文纵览语言模型推理能力
Multimodality in Medical Domain：一文速览医学多模态进展