72 Matching Annotations
  1. Mar 2019
    1. A Network-based End-to-End Trainable Task-oriented Dialogue System

      这个end-to-end的系统,在意图识别的阶段用的是cnn+LSTM 在状态管理(belief state tracking)也用的LSTM,在policy的时候自定义了一套算法,将前面的几个输出向量做了个线性模型,输出。

    1. Neural Approaches to Conversational AI

      Question Answering, Task-Oriented Dialogues and Social Chatbots

      The present paper surveys neural approaches to conversational AI that have beendeveloped in the last few years. We group conversational systems into three cat-egories: (1) question answering agents, (2) task-oriented dialogue agents, and(3) chatbots. For each category, we present a review of state-of-the-art neuralapproaches, draw the connection between them and traditional approaches, anddiscuss the progress that has been made and challenges still being faced, usingspecific systems and models as case studies

    1. Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling

      用一个模型来解决两个不同类型的问题,intent detect是分类,填槽是序列标注。都用基于attention机制的RNN来搞定了

    1. We present a general solution towards building task-orienteddialogue systems for online shopping, aiming to assist on-line customers in completing various purchase-related tasks,such as searching products and answering questions, in a nat-ural language conversation manner. As a pioneering work, weshow what & how existing natural language processing tech-niques, data resources, and crowdsourcing can be leveragedto build such task-oriented dialogue systems for E-commerceusage. To demonstrate its effectiveness, we integrate our sys-tem into a mobile online shopping application. To the bestof our knowledge, this is the first time that an dialogue sys-tem in Chinese is practically used in online shopping scenariowith millions of real consumers. Interesting and insightful ob-servations are shown in the experimental part, based on theanalysis of human-bot conversation log. Several current chal-lenges are also pointed out as our future directions



      M = (I, C, A)

      I是intent,C是product category, A是商品attribute。 M是根据用户Query得到的信息的表示。

      意图分类:PhraseLDA 1000个topic

      产品分类: a CNN-based approach that resembles (Huang et al. 2013)and (Shen et al. 2014

    2. Main actions that areconsidered in the online shopping scenario include


      • Recommendation
      • Comparison
      • Opinion Summary
      • Question Answering
      • Proactive Questioning
      • Chit-chat
    3. To deal with the problem we mentioned, our work focuson using three kinds of data resources that are common tomost E-commerce web service provider or easily crawledfrom webs, including: (i) product knowledge base, which isprovided by the E-commerce partner and contains structuredproduct information; (ii) search log, which is closely linkedwith products, natural language queries and user selectionbehaviors (mouse click); (iii) community sites, where userpost their intents in natural language and can be used to minepurchase-related intents and paraphrases of product-relatedterms. Besides, we show that crowd sourcing is necessary tobuild such AI bot


      • 1 结构化商品信息
      • 2 用户的搜索日志
      • 3 社区网站,挖掘购买意图和产品相关的词
    1. Retrieval-based MethodsRetrieval-based methods choose a response from candidateresponses. The key to retrieval-based methods is message-response matching. Matching algorithms have to overcomesemantic gaps between messages and responses [28].


      B. Hu, Z. Lu, H. Li, and Q. Chen. Convolutional neu-ral network architectures for matching natural lan-guage sentences. InAdvances in neural informationprocessing systems, pages 2042–2050, 2014.

      单轮的匹配 match(X,Y) = X^TAy

      X:message的向量表示, y:回复的向量表示。

      H. Wang, Z. Lu, H. Li, and E. Chen. A dataset for re-search on short-text conversations. InProceedings ofthe 2013 Conference on Empirical Methods in NaturalLanguage Processing, pages 935–945, Seattle, Wash-ington, USA, October 2013. Association for Compu-tational Linguistics

      Z. Lu and H. Li. A deep architecture for matchingshort texts. InInternational Conference on Neural In-formation Processing Systems, pages 1367–1375, 2013.

      B. Hu, Z. Lu, H. Li, and Q. Chen. Convolutional neu-ral network architectures for matching natural lan-guage sentences. InAdvances in neural informationprocessing systems, pages 2042–2050, 2014

      M. Wang, Z. Lu, H. Li, and Q. Liu. Syntax-based deepmatching of short texts.InIJCAI, 03 2015

      Y. Wu, W. Wu, Z. Li, and M. Zhou. Topic augmentedneural network for short text conversation.CoRR,2016


    2. 2.1.3 Policy learning

      策略学习 基于前面state tracker的状态表示,策略学习(policy learning)是来生成下一个可用的系统行动。无论是监督学习或者强化学习都可以用来优化策略学习。 H. Cuayhuitl, S. Keizer, and O. Lemon. Strategic di-alogue management via deep reinforcement learning.arxiv.org, 2015.

      通常都用一个基于规则的agent来初始化系统。 Z. Yan, N. Duan, P. Chen, M. Zhou, J. Zhou, andZ. Li. Building task-oriented dialogue systems for on-line shopping. InAAAI Conference on Artificial Intel-ligence, 2017

      然后用监督学习来基于规则生成的规则来学习。Building task-oriented dialogue systems for on-line shopping. 强化学习,Strategic di-alogue management via deep reinforcement learning.结果据说比很多系统,rule based,superviesed都好

  2. Feb 2019
    1. To overcome this issue, weexplore data generation using templates and terminologies and data augmentationapproaches. Namely, we report our experiments using paraphrasing and wordrepresentations learned on a large EHR corpus with Fasttext and ELMo, to learn aNLU model without any available dataset. We evaluate on a NLU task of naturallanguage queries in EHRs divided in slot-filling and intent classification sub-tasks.On the slot-filling task, we obtain a F-score of 0.76 with the ELMo representation;and on the classification task, a mean F-score of 0.71. Our results show that thismethod could be used to develop a baseline system



    2. Natural language understanding for task oriented dialog in the biomedical domain in a low resources context

    1. PyDial: A Multi-domain Statistical Dialogue System Toolkit


      其总的架构包含Sematic Decode,Belief Tracker,Policy Reply System,Language generator. 整体来说整个系统都支持了基于规则的判断过程,也融合了模型的支持。源码值得一看的。

  3. www.iro.umontreal.ca www.iro.umontreal.ca
    1. Using Recurrent Neural Networks for Slot Filling in Spoken Language Understanding

    2. Using Recurrent Neural Networksfor Slot Filling in Spoken Language Understanding

    1. ntegrating User and Agent Models: A Deep Task-Oriented Dialogue System Weiyan Wang, Yuxiang WU, Yu Zhang, Zhongqi Lu, Kaixiang Mo, Qiang Yang (Submitted on 10 Nov 2017) Task-oriented dialogue systems can efficiently serve a large number of customers and relieve people from tedious works. However, existing task-oriented dialogue systems depend on handcrafted actions and states or extra semantic labels, which sometimes degrades user experience despite the intensive human intervention. Moreover, current user simulators have limited expressive ability so that deep reinforcement Seq2Seq models have to rely on selfplay and only work in some special cases. To address those problems, we propose a uSer and Agent Model IntegrAtion (SAMIA) framework inspired by an observation that the roles of the user and agent models are asymmetric. Firstly, this SAMIA framework model the user model as a Seq2Seq learning problem instead of ranking or designing rules. Then the built user model is used as a leverage to train the agent model by deep reinforcement learning. In the test phase, the output of the agent model is filtered by the user model to enhance the stability and robustness. Experiments on a real-world coffee ordering dataset verify the effectiveness of the proposed SAMIA framework.

    1. Deep Reinforcement Learning for Dialogue Generation

      Recent neural models of dialogue generationoffer great promise for generating responsesfor conversational agents, but tend to be short-sighted, predicting utterances one at a timewhile ignoring their influence on future out-comes. Modeling the future direction of a di-alogue is crucial to generating coherent, inter-esting dialogues, a need which led traditionalNLP models of dialogue to draw on reinforce-ment learning. In this paper, we show how tointegrate these goals, applying deep reinforce-ment learning to model future reward in chat-bot dialogue. The model simulates dialoguesbetween two virtual agents, using policy gradi-ent methods to reward sequences that displaythree useful conversational properties: infor-mativity, coherence, and ease of answering (re-lated to forward-looking function). We evalu-ate our model on diversity, length as well aswith human judges, showing that the proposedalgorithm generates more interactive responsesand manages to foster a more sustained conver-sation in dialogue simulation. This work marksa first step towards learning a neural conversa-tional model based on the long-term success ofdialogues.

    1. Dialog System & Technology Challenge 6 Overview of Track 1 - End-to-End Goal-Oriented Dialog learning

      End-to-end dialog learning is an important research subject inthe domain of conversational systems. The primary task consistsin learning a dialog policy from transactional dialogs of a givendomain. In this context, usable datasets are needed to evaluatelearning approaches, yet remain scarce. For this challenge, atransaction dialog dataset has been produced using a dialogsimulation framework developed and released by Facebook AIResearch. Overall, nine teams participated in the challenge. Inthis report, we describe the task and the dataset. Then, we specifythe evaluation metrics for the challenge. Finally, the results ofthe submitted runs of the participants are detailed.

    1. Task-oriented dialog systems 要观看此视频,请启用 JavaScript 并考虑升级到 支持 HTML5 视频 的 Web 浏览器 Video Player is loading.Play VideoPlayMuteLoaded: 0%Progress: 0%Subtitlessubtitles off, selected英语(English)Quality360p720p540p360p, selectedFullscreenThis is a modal window.Beginning of dialog window. Escape will cancel and close the window.TextColorWhiteBlackRedGreenBlueYellowMagentaCyanTransparencyOpaqueSemi-TransparentBackgroundColorBlackWhiteRedGreenBlueYellowMagentaCyanTransparencyOpaqueSemi-TransparentTransparentWindowColorBlackWhiteRedGreenBlueYellowMagentaCyanTransparencyTransparentSemi-TransparentOpaqueFont Size50%75%100%125%150%175%200%300%400%Text Edge StyleNoneRaisedDepressedUniformDropshadowFont FamilyProportional Sans-SerifMonospace Sans-SerifProportional SerifMonospace SerifCasualScriptSmall CapsReset restore all settings to the default valuesDoneClose Modal DialogEnd of dialog window.

      coursera 对话系统课程

    1. 京东的机器人架构

      Existing solutions to task-oriented dia-logue systems follow pipeline designswhich introduce architectural complex-ity and fragility. We propose a novel,holistic, extendable framework based ona single sequence-to-sequence (seq2seq)model which can be optimized with su-pervised or reinforcement learning. Akey contribution is that we design textspans namedbelief spansto track dia-logue believes, allowing task-oriented dia-logue systems to be modeled in a seq2seqway. Based on this, we propose a sim-plisticTwo Stage CopyNetinstantiationwhich demonstrates good scalability: sig-nificantly reducing model complexity interms of number of parameters and train-ing time by an order of magnitude. Itsignificantly outperforms state-of-the-artpipeline-based methods on two datasetsand retains a satisfactory entity match rateon out-of-vocabulary (OOV) cases wherepipeline-designed competitors totally fail

    1. A Survey on Dialogue Systems:Recent Advances and New Frontiers

      Dialogue systems have attracted more and more attention.Recent advances on dialogue systems are overwhelminglycontributed by deep learning techniques, which have beenemployed to enhance a wide range of big data applicationssuch as computer vision, natural language processing, andrecommender systems. For dialogue systems, deep learningcan leverage a massive amount of data to learn meaningfulfeature representations and response generation strategies,while requiring a minimum amount of hand-crafting. In thisarticle, we give an overview to these recent advances on di-alogue systems from various perspectives and discuss somepossible research directions. In particular, we generally di-vide existing dialogue systems into task-oriented and non-task-oriented models, then detail how deep learning tech-niques help them with representative algorithms and ?nallydiscuss some appealing research directions that can bringthe dialogue system research into a new frontier.


      We introduce end-to-end neural network based models for simulat-ing users of task-oriented dialogue systems. User simulation in di-alogue systems is crucial from two different perspectives: (i) auto-matic evaluation of different dialogue models, and (ii) training task-oriented dialogue systems. We design a hierarchical sequence-to-sequence model that first encodes the initial user goal and systemturns into fixed length representations using Recurrent Neural Net-works (RNN). It then encodes the dialogue history using anotherRNN layer. At each turn, user responses are decoded from the hid-den representations of the dialogue level RNN. This hierarchical usersimulator (HUS) approach allows the model to capture undiscov-ered parts of the user goal without the need of an explicit dialoguestate tracking. We further develop several variants by utilizing a la-tent variable model to inject random variations into user responses topromote diversity in simulated user responses and a novel goal regu-larization mechanism to penalize divergence of user responses fromthe initial user goal. We evaluate the proposed models on movieticket booking domain by systematically interacting each user sim-ulator with various dialogue system policies trained with differentobjectives and users.

    1. 腾讯的数据集

    2. A Manually Annotated Chinese Corpus forNon-task-oriented Dialogue SystemsJing Li, Yan Song, Haisong Zhang, Shuming ShiTencent AI Labfameliajli,clksong,hansonzhang,shumingshig@tencent.comAbstractThis paper presents a large-scale corpusfor non-task-oriented dialogue responseselection, which contains over27K dis-tinct prompts more than82K responsescollected from social media.1To anno-tate this corpus, we define a5-grade rat-ing scheme:bad,mediocre,acceptable,good, andexcellent, according to the rel-evance, coherence, informativeness, inter-estingness, and the potential to move aconversation forward. To test the valid-ity and usefulness of the produced cor-pus, we compare various unsupervised andsupervised models for response selection.Experimental results confirm that the pro-posed corpus is helpful in training re-sponse selection models

      A Manually Annotated Chinese Corpus forNon-task-oriented Dialogue SystemsJing Li, Yan Song, Haisong Zhang, Shuming ShiTencent AI Labfameliajli,clksong,hansonzhang,shumingshig@tencent.comAbstractThis paper presents a large-scale corpusfor non-task-oriented dialogue responseselection, which contains over27K dis-tinct prompts more than82K responsescollected from social media.1To anno-tate this corpus, we define a5-grade rat-ing scheme:bad,mediocre,acceptable,good, andexcellent, according to the rel-evance, coherence, informativeness, inter-estingness, and the potential to move aconversation forward. To test the valid-ity and usefulness of the produced cor-pus, we compare various unsupervised andsupervised models for response selection.Experimental results confirm that the pro-posed corpus is helpful in training re-sponse selection models

    1. A Deep Reinforcement Learning Chatbot

      开放领域的对话机器人 是亚马逊Alexa比赛的结果 不同于历史的大部分基于规则对话系统,该机器人采用以统计机器学习方法为主的模型来构建机器人。



      • 1.各个响应模型先生成各自的回复
      • 2.如果有个“优先”回复,直接返回
      • 3.如果没有“优先”回复,则根据model selection 策略进行选择




    2. To generate a response, the dialogue manager follows a three-stepprocedure. First, it uses all response models to generate a set of candidate responses. Second, if thereexists apriorityresponse in the set of candidate responses (i.e. a response which takes precedenceover other responses), this response will be returned by the system.5For example, for the question"What is your name?", the response"I am an Alexa Prize socialbot"is a priority response. Third, ifthere are nopriorityresponses, the response is selected by themodel selection policy. For example,themodel selection policymay select a response by scoring all candidate responses and picking thehighest-scored response.


      • 1.各个响应模型先生成各自的回复
      • 2.如果有个“优先”回复,直接返回
      • 3.如果没有“优先”回复,则根据model selection 策略进行选择
    3. There are 22 response models in the system, including retrieval-based neural networks, generation-based neural networks, knowledge base question answering systems and template-based systems.Examples of candidate model responses are shown in Tabl


    4. Early work on dialogue systems (Weizenbaum 1966, Colby 1981, Aust et al. 1995, McGlashan et al.1992, Simpson & Eraser 1993) were based mainly on states and rules hand-crafted by human experts.Modern dialogue systems typically follow a hybrid architecture, combining hand-crafted states andrules with statistical machine learning algorithms (Suendermann-Oeft et al. 2015, Jurˇcíˇcek et al.2014, Bohus et al. 2007, Williams 2011).


    5. We present MILABOT: a deep reinforcement learning chatbot developed by theMontreal Institute for Learning Algorithms (MILA) for the Amazon Alexa Prizecompetition. MILABOT is capable of conversing with humans on popular smalltalk topics through both speech and text. The system consists of an ensemble ofnatural language generation and retrieval models, including template-based models,bag-of-words models, sequence-to-sequence neural network and latent variableneural network models. By applying reinforcement learning to crowdsourced dataand real-world user interactions, the system has been trained to select an appropriateresponse from the models in its ensemble. The system has been evaluated throughA/B testing with real-world users, where it performed significantly better thanmany competing systems. Due to its machine learning architecture, the system islikely to improve with additional data
    1. The Design and Implementation of XiaoIce,an Empathetic Social Chatbot

      主要的设计目标是维持和人的长期情感联系。平均CPS 23.智商+情商。

      empathetic computingframework共情计算框架

      Yang Cai. Empathic computing. InAmbient Intelligence in Everyday Life, pages 67–85.Springer Pascale Fung, Dario Bertero, Yan Wan, Anik Dey, Ricky Ho

      Yin Chan, Farhad Bin Siddique,Yang Yang, Chien-Sheng Wu, and Ruixi Lin. Towards empathetic human-robot interactions.CoRR, abs/1605.04072, 2016

      If a person enjoys its companionship (via conversation), we can call the machine “empathetic”

      IQ知识和记忆模型,图片和自然语言理解,推理,生成和预测 EQ情绪识别,理解情感需求

    2. The Design and Implementation of XiaoIce,an Empathetic Social Chatbot


    1. In this paper, we describe an architecture for conversationalsystems that enables human-like performance along severalimportant dimensions. First, interpretation is incremental,multi-level, and involves both general and task- and domain-specific knowledge. Second, generation is also incremental,proceeds in parallel with interpretation, and accounts forphenomena such as turn-taking, grounding and interrup-tions. Finally, the overall behavior of the system in thetask at hand is determined by the (incremental) results ofinterpretation, the persistent goals and obligations of thesystem, and exogenous events of which it becomes aware. Asa practical matter, the architecture supports a separation ofresponsibilities that enhances portability to new tasks anddomains
    1. ecent advances of deep learning have inspiredmany applications of neural models to dialoguesystems. Wen et al. (2017) and Bordes et al.(2017) introduced a network-based end-to-endtrainable task-oriented dialogue system, whichtreated dialogue system learning as the problemof learning a mapping from dialogue histories tosystem responses, and applied an encoder-decodermodel to train the whole system



    2. Antoine Bordes, Y-Lan Boureau, and Jason Weston.2017. Learning end-to-end goal-oriented dialog. InProceedings of ICLR
    3. End-to-End Task-Completion Neural Dialogue Systems

      One of the major drawbacks of modu-larized task-completion dialogue systemsis that each module is trained individu-ally, which presents several challenges.For example, downstream modules are af-fected by earlier modules, and the per-formance of the entire system is not ro-bust to the accumulated errors. This pa-per presents a novel end-to-end learningframework for task-completion dialoguesystems to tackle such issues. Our neu-ral dialogue system can directly interactwith a structured database to assist usersin accessing information and accomplish-ing certain tasks. The reinforcement learn-ing based dialogue manager offers robustcapabilities to handle noises caused byother components of the dialogue system.Our experiments in a movie-ticket book-ing domain show that our end-to-end sys-tem not only outperforms modularized di-alogue system baselines for both objectiveand subjective evaluation, but also is ro-bust to noises as demonstrated by severalsystematic experiments with different er-ror granularity and rates specific to the lan-guage understanding module1

      The source code is available at: https://github.com/MiuLab/TC-Bot

    1. AbstractWe propose Neural Responding Ma-chine (NRM), a neural network-based re-sponse generator for Short-Text Conver-sation. NRM takes the general encoder-decoder framework: it formalizes the gen-eration of response as a decoding processbased on the latent representation of the in-put text, while both encoding and decod-ing are realized with recurrent neural net-works (RNN). The NRM is trained witha large amount of one-round conversationdata collected from a microblogging ser-vice. Empirical study shows that NRMcan generate grammatically correct andcontent-wise appropriate responses to over75% of the input text, outperforming state-of-the-arts in the same setting, includingretrieval-based and SMT-based models


    1. DocChat: An Information Retrieval Approach for Chatbot EnginesUsing Unstructured Documents



      最有用的是sentence level的特征。

    2. information retrieval approach for chat-bot engines that can leverage unstructureddocuments, instead of Q-R pairs, to re-spond to utterances. A learning to rankmodel with features designed at differentlevels of granularity is proposed to mea-sure the relevance between utterances andresponses directly.


    3. a much simplified task, short textconversation (STC) in which the responseRis ashort text and only depends on the last user utter-anceQ.

      STC 只依赖用户上一句输入的短文本对话过程

    1. specch act classification的难点

      • 1 同一个词在不同语境中有不同的意思
      • 2 说话意图的识别本身也比较难做

      歧义,特异,数据源 ambiguity,specificity,source

  4. Oct 2018
  5. Jun 2018
  6. Feb 2018
  7. Feb 2017
    1. Even creepier, a startup created an AI-powered memorial chatbot: software that can learn everything about you from your chat logs, and then allow your friends to chat with your digital-self after you die.

      Does this sound familiar?