Dialogue System Survey

布满荆棘的人生 2022-04-22 12:22 118阅读 0赞

# Dialogue System Survey #

*  Task-oriented
 *  Non-task-oriented
 *  Discussion

![IMG\_0391.JPG][IMG_0391.JPG]

## Task-oriented dialogue system ##

目的：帮助用户完成某些特定任务，比如查找产品、客服等

特点：通常需要在外部知识库上进行查询

### Pipeline Methods ###

![图片 1.png][1.png]

基于管道的面向任务的对话系统的四个组成部分：

*  自然语言理解（NLU）：将用户的语义解析为预定义的语义槽，并进行意图检测和槽填充（NER）
    
    ![图片 2.png][2.png]
 *  对话状态跟踪器： 管理每轮输入以及对话历史并输出当前的对话状态（手工规则/DL）
 *  对话策略学习：根据当前的对话状态学习下一步行动（Supervised learning、RL）
 *  自然语言生成（NLG）：将选定的动作转化为自然语言表达

缺点：1. 用户反馈难以传达到上游模块  
2. 模块间相互依赖

> NLU  
> L. Deng, G. Tur, X. He, and D. Hakkani-Tur. Use of kernel deep convex networks and end-to-end learning for spoken language understanding. In Spoken Lan- guage Technology Workshop (SLT), 2012 IEEE, pages 210–215. IEEE, 2012.  
> P.-S. Huang, X. He, J. Gao, L. Deng, A. Acero, and L. Heck. Learning deep structured semantic models for web search using clickthrough data. In Proceedings of the 22nd ACM international conference on Confer- ence on information & knowledge management, pages 2333–2338. ACM, 2013.
> 
> Dialog state tracking  
> D. Goddeau, H. Meng, J. Polifroni, S. Seneff, and S. Busayapongchai. A form-based dialogue manager for spoken language applications. In Spoken Language, 1996. ICSLP 96. Proceedings., Fourth International Conference on, volume 2, pages 701–704. IEEE, 1996.  
> N. Mrkˇ si´ c, D.Ó Séaghdha, B. Thomson, M. Gasic, P.- H. Su, D. Vandyke, T.-H. Wen, and S. Young. Multi- domain dialog state tracking using recurrent neural networks. In Proceedings of the 53rd Annual Meet- ing of the Association for Computational Linguistics and the 7th International Joint Conference on Nat- ural Language Processing (Volume 2: Short Papers), pages 794–799, Beijing, China, July 2015. Association for Computational Linguistics.
> 
> Policy learing  
> H. Cuayhuitl, S. Keizer, and O. Lemon. Strategic di- alogue management via deep reinforcement [learning.arxiv.org][], 2015.
> 
> NLG  
> A.Stent, M. Marge, and M. Singhai. Evaluating eval- uation methods for generation in the presence of vari- ation. In International Conference on Computational Linguistics and Intelligent Text Processing, pages 341–351, 2005.  
> H. Zhou, M. Huang, and X. Zhu. Context-aware nat- ural language generation for spoken dialogue [systems.In][] COLING, pages 2032–2041, 2016.

### End-to-End Methods ###

![图片 5.png][5.png]

一个可以与结构化外部数据交互的模块

> T.-H. Wen, D. Vandyke, N. Mrkˇ si´ c, M. Gasic, L. M.Rojas Barahona, P.-H. Su, S. Ultes, and S. Young. A network-based end-to-end trainable task-oriented di- alogue system. In Proceedings of the 15th Conference of the European Chapter of the Association for Com- putational Linguistics: Volume 1, Long Papers, pages 438–449, Valencia, Spain, April 2017. Association for Computational Linguistics.  
> A. Bordes, Y. L. Boureau, and J. Weston. Learning end-to-end goal-oriented dialog. In ICLR, 2017.

## Non-task-oriented dialogue system (chatbot) ##

目的：在开放领域生成回复

### Neural Generative Models ###

![图片 4.png][4.png]

Seq2seq+

*  Context
    
     *  层次RNN通常优于非层次RNN
     *  对于上下文信息，NN倾向于产生更长，更有意义和更多样的回复
 *  Diversity
    
     *  找到并设置一个更好的目标函数
     *  增加模型的复杂度
 *  Topic & Personality
 *  Outside Knowledge Base
    
     *  Memory Network
 *  Interactive Dialogue learning
    
     *  GAN
     *  开发者定义reward，而不是通过Seq2seq的目标函数来学习策略

> Context  
> A.Sordoni, M. Galley, M. Auli, C. Brockett, Y. Ji, M. Mitchell, J.-Y. Nie, J. Gao, and B. Dolan. A neu- ral network approach to context-sensitive generation of conversational responses. In Proceedings of the 2015 Conference of the North American Chapter of the As- sociation for Computational Linguistics: Human Lan- guage Technologies, pages 196–205, Denver, Colorado, May–June 2015. Association for Computational Lin- guistics.
> 
> Diversity  
> J. Li, M. Galley, C. Brockett, J. Gao, and B. Dolan. A diversity-promoting objective function for neural con- versation models. In Proceedings of the 2016 Confer- ence of the North American Chapter of the Associa- tion for Computational Linguistics: Human Language Technologies, pages 110–119, San Diego, California, June 2016. Association for Computational Linguistics.  
> K. Cao and S. Clark. Latent variable dialogue models and their diversity. In Proceedings of the 15th Confer- ence of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, pages 182–187, Valencia, Spain, April 2017. Associa- tion for Computational Linguistics.
> 
> Topic & Personality  
> C. Xing, W. Wu, Y. Wu, J. Liu, Y. Huang, M. Zhou, and W.-Y. Ma. Topic aware neural response genera- tion. In AAAI Conference on Artificial Intelligence, 2017.  
> H. Zhou, M. Huang, T. Zhang, X. Zhu, and B. Liu.Emotional chatting machine: Emotional conversation generation with internal and external memory. arXiv preprint arXiv:1704.01074, 2017.  
> M. Qiu, F.-L. Li, S. Wang, X. Gao, Y. Chen, W. Zhao, H. Chen, J. Huang, and W. Chu. Alime chat: A se- quence to sequence and rerank based chatbot engine.Œ  
> In Proceedings of the 55th Annual Meeting of the As- sociation for Computational Linguistics (Volume 2: Short Papers), volume 2, pages 498–503, 2017.
> 
> Outside Knowledge Base  
> M.Ghazvininejad, C.Brockett, M.-W.Chang, B. Dolan, J. Gao, W.-t. Yih, and M. Galley. A knowledge-grounded neural conversation model. arXiv preprint arXiv:1702.01932, 2017.  
> P. Vougiouklis, J. Hare, and E. Simperl. A neural net- work approach for knowledge-driven response gener- ation. In Proceedings of COLING 2016, the 26th In- ternational Conference on Computational Linguistics: Technical Papers, pages 3370–3380, Osaka, Japan, De- cember 2016. The COLING 2016 Organizing Commit- tee.
> 
> Interactive Dialogue learning  
> R. J. Williams. Simple statistical gradient-following al- gorithms for connectionist reinforcement learning. Ma- chine learning, 8(3-4):229–256, 1992.  
> M. Lewis, D. Yarats, Y. Dauphin, D. Parikh, and D. Batra. Deal or no deal? end-to-end learning of ne- gotiation dialogues. In Proceedings of the 2017 Con- ference on Empirical Methods in Natural Language Processing, pages 2433–2443, Copenhagen, Denmark, September 2017. Association for Computational Lin- guistics

Evaluation

*  Human-generated supervised signals
 *  Automatically metrics
    
     *  word overlap (BLEU、ROUGE等MT的评估方法，可能不太适用于NRG）
     *  计算 embedding的距离（直接相加求平均、先取绝对值再求平均和贪婪匹配）
     *  衡量多样性（主要取决于 distinct-ngram 的数量和 entropy 值的大小）
     *  Turing test（根据区分机器回复与人类回复的容易程度打分）

> Human-generated supervised signals  
> M. A. Walker, D. J. Litman, C. A. Kamm, and A. Abella. Paradise: A framework for evaluating spo- ken dialogue agents. In Proceedings of the eighth con- ference on European chapter of the Association for Computational Linguistics, pages 271–280. Associa- tion for Computational Linguistics, 1997.
> 
> Automatically metrics  
> R. Artstein, S. Gandhe, J. Gerten, A. Leuski, and D. Traum. Semi-formal evaluation of conversational characters. pages 22–35, 2009.  
> C.-W. Liu, R. Lowe, I. Serban, M. Noseworthy, L. Charlin, and J. Pineau. How not to evaluate your dialogue system: An empirical study of unsupervised evaluation metrics for dialogue response generation. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2122–2132, Austin, Texas, November 2016. Association for Computational Linguistics.  
> R. Lowe, M. Noseworthy, I. V. Serban, N. Angelard- Gontier, Y. Bengio, and J. Pineau. Towards an auto- matic turing test: Learning to evaluate dialogue re- sponses. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Vol- ume 1: Long Papers), pages 1116–1126, Vancouver, Canada, July 2017. Association for Computational Linguistics.

### Retrieval-based Methods ###

*  Single-turn Response Matching
    
    ![图片 6.png][6.png]
    
     *  CNN学习消息和响应的表示，或直接学习两个句子的相互作用表示，然后用多层感知器来计算匹配的分数。
 *  Multi-turn Response Matching
    
    ![图片 3.png][3.png]
    
     *  明确上下文的关键信息（关键词，关键短语或关键句）
     *  在上下文中模拟多轮对话间的关系
 *  Hybrid Methods
    
     *  被抽取的候选对象和原始消息一起被输入到基于RNN的回复生成器中

> Single-turn Response Matching  
> H. Wang, Z. Lu, H. Li, and E. Chen. A dataset for re- search on short-text conversations. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 935–945, Seattle, Wash- ington, USA, October 2013. Association for Compu- tational Linguistics.
> 
> Multi-turn Response Matching  
> R. Lowe, N. Pow, I. Serban, and J. Pineau. The ubuntu dialogue corpus: A large dataset for research in unstructured multi-turn dialogue systems. In Pro- ceedings of the 16th Annual Meeting of the Special In- terest Group on Discourse and Dialogue, pages 285–294, Prague, Czech Republic, September 2015. Asso- ciation for Computational Linguistics.
> 
> Hybrid Methods  
> Y. Song, R. Yan, X. Li, D. Zhao, and M. Zhang.Two are better than one: An ensemble of retrieval- and generation-based dialog systems. arXiv preprint arXiv:1610.07149, 2016.  
> M. Qiu, F.-L. Li, S. Wang, X. Gao, Y. Chen, W. Zhao, H. Chen, J. Huang, and W. Chu. Alime chat: A se- quence to sequence and rerank based chatbot engine. In Proceedings of the 55th Annual Meeting of the As- sociation for Computational Linguistics (Volume 2: Short Papers), volume 2, pages 498–503, 2017.

### Discussion ###

*  Active learning
    
    快速适应。虽然端到端模型越来越引起研究者的重视，我们仍然需要在实际工程中依靠传统的管道（pipeline）方法，特别是在一些新的领域，特定领域对话数据的收集和对话系统的构建是比较困难的。未来的趋势是对话模型有能力从与人的交互中主动去学习。
 *  Deep Understanding
    
    深度理解。现阶段基于神经网络的对话系统极大地依赖于大量标注好的数据，结构化的知识库以及对话语料数据。在某种意义上产生的回复仍然缺乏多样性，有时并没有太多的意义，因此对话系统必须能够更加有效地深度理解语言和真实世界。
 *  Privacy Protection
    
    隐私保护。目前广泛应用的对话系统服务于越来越多的人。很有必要注意到的事实是我们使用的是同一个对话助手。通过互动、理解和推理的学习能力，对话助手可以无意中隐蔽地存储一些较为敏感的信息。因此，在构建更好的对话机制时，保护用户的隐私是非常重要的。

> N. Papernot, M. Abadi,Ú. Erlingsson, I. Goodfellow, and K. Talwar. Semi-supervised knowledge transfer for deep learning from private training data. ICLR, 2017.

## Reference ##

[A Survey on Dialogue Systems: Recent Advances and New Frontiers][A Survey on Dialogue Systems_ Recent Advances and New Frontiers]

[对话系统调查：最新进展与新前沿][Link 1]

[IMG_0391.JPG]: https://i.loli.net/2018/11/05/5bdf9bcb71ce0.jpg
[1.png]: https://i.loli.net/2018/11/05/5bdf8e3e961d8.png
[2.png]: https://i.loli.net/2018/11/05/5bdf8ea7e38ce.png
[learning.arxiv.org]: http://learning.arxiv.org
[systems.In]: http://systems.In
[5.png]: https://i.loli.net/2018/11/05/5bdf8ea827fc7.png
[4.png]: https://i.loli.net/2018/11/05/5bdf8ea8563b3.png
[6.png]: https://i.loli.net/2018/11/05/5bdf8ea892b01.png
[3.png]: https://i.loli.net/2018/11/05/5bdf8ea8a29e4.png
[A Survey on Dialogue Systems_ Recent Advances and New Frontiers]: https://arxiv.org/pdf/1711.01731.pdf
[Link 1]: https://blog.csdn.net/angus_monroe/article/details/82048989