![Intelligent Interaction Technical Practice -Task Bot
RL for task-bot
related-work from MSR
References
Williams J D, Zweig G. End-to-end LSTM-based dialog control optimized with
supervised and reinforcement learning[J]. 2016.](https://image.slidesharecdn.com/11haiqingchen-171206021110/85/AliMe-Bot-Platform-Technical-Practice-Alibaba-s-Personal-Intelligent-Assistant-in-the-E-commerce-Field-24-320.jpg)
![Intelligent Interaction Technical Practice -Task Bot
DRL for task-bot
related-work【From Cambrige】
References
Wen T H, Vandyke D, Mrksic N, et al. A Network-based End-to-End
Trainable Task-oriented Dialogue System[J]. 2016.
task generation
action generation
dialogue managerment
belief-tracking
intent
representation
context
slot-extraction pattern-extration
word seg ner
task generation
reinforcement
learning
data
preprocess](https://image.slidesharecdn.com/11haiqingchen-171206021110/85/AliMe-Bot-Platform-Technical-Practice-Alibaba-s-Personal-Intelligent-Assistant-in-the-E-commerce-Field-25-320.jpg)
AliMe Bot Platform Technical Practice - Alibaba`s Personal Intelligent Assistant in the E-commerce Field
- 1. AliMe Bot Platform Technical Practice - Alibaba’s Personal Intelligent Assistant Alibaba Group – Intelligent Innovation Division ——Haiqing Chen(Hunter)
- 2. Outline AliMe Bot Platform Introduction Intelligent Interaction Technical Practice Look Forward to The Future
- 3. Outline AliMe Bot Platform Introduction Intelligent Interaction Technical Practice Look Forward to The Future
- 4. AliMe Bot Platform Introduction Customer Service Self Support Outsource Cloud Service Customer Service Self Support Outsource Cloud Service Chatbot Platform Platf orm 3rd party AlibabaMerchantEnterprise Upgrading of service: From manpower to intelligence plus manpower Upgrading of field: From Ali inside to merchants and enterprises
- 5. AliMe Bot Platform Introduction AliMe bot platform is a complete set of intelligent human-computer interaction solution. • Face to three domains solutions: • AliMe Assistant: Platform solution for Alibaba business • TiMi: Platform solution for Merchants of Taobao • Bee Bot: Platform solution for enterprises, ISV etc. • Supporting two types of infrastructures: • SaaS: Front to end solution, a whole chatbot system including chat UI • PaaS: Supply AI interfaces capability for developers to help build their systems
- 6. Configuration Platform Chatbot QA Platform Smart Knowledge Base AI Boost Chatbot Application Platform AliMe Assistant Platform Different Domains of AliMe Assistants TiMi Alibaba Merchant Taobao Tmall SaaS Enterprise Super AliMe Assistant Bee Bot Platform Wang Xiang Travel Second- hand Other B2B Logistics Youku Other SaaS PaaS Qianniu Platform DingTalk AliCloud IOT SaaS Bot Framework Algorithm Component Platform Data Platform Oversea SaaS or PaaS AliMe Bot Platform Introduction
- 7. AliMe Bot Platform Introduction AliMe Assistant • Covered Areas: • Customer Service • Shopping Guide • General Assistant Application • Chit Chat • Activity Operation • ……
- 8. AliMe Bot Platform Introduction TiMi • Features: • General ontology model for all categories: logistics field • Industry ontology model: mobile phone, clothing etc. • General QA model
- 9. AliMe Bot Platform Introduction Bee Bot Platform • Features: • This is a whole solution for 3rd party, i.e. enterprise, ISV etc. • Smart knowledge base to fullfill knowledges
- 10. Outline AliMe Bot Platform Introduction Intelligent Interaction Technical Practice Look Forward to The Future
- 11. Intelligent Interaction Technical Practice AliMe Assistant TiMi Bee Bot Platform PaaS Service Layer Dialog Management System Matching Model Bot Framework Algorithm Component Platform Configuration Platform Knowledge Base AI Boost AliMe Bot Platform Application Architecture
- 12. Intelligent Interaction Technical Practice Intent Routing Customer Service Shopping Guide Logistics Field Chit Chat others Image Recognition Muti-turn Interaction ASR Semantic Recognition User Profile Muti-model Interaction Recommendation Knowledge(QA pairs, Ontology, Knowledge Graph) Data AnalyticsData MiningML/DL Training Routing Layer Service Layer Model Layer Data Layer AliMe Bot Platform Algorithm Model Architecture
- 13. Intelligent Interaction Technical Practice Query+Context Intention Recognition Dialog Management System QA Bot Task Bot Chat Bot Kowledge Graph IR Slot Filling Bot Framework DRL IR+LSTM
- 14. Intelligent Interaction Technical Practice Intention Identification Process Query Segmentation, POS Tagging, NER Intention Classification Dialog Management Semantic Representation Intention Attribute Extraction Domain Model Context Model
- 15. Intelligent Interaction Technical Practice Intention Identification Classification • Traditional machine Learning methods : • Supervised multiply classification • Supervised multiply binary classification • E.g. Bayes, KNN, SVM, Logistic Regression etc. • Deep Learning: • Combined user profile or behavior to build deep learning prediction model • E.g. CNN, DNN, LSTM etc.
- 16. Intelligent Interaction Technical Practice Enrich Question with Profile, Behavior and Context Question User Behavior User Profile Query+ Context DL Model
- 17. Intelligent Interaction Technical Practice Intention Identification with Deep Learning N V Question (BOW/RNN/CNN) Behavior, Profile, Context L Embedding Softmax N V Question (BOW/RNN/CNN) Behavior, Profile, Context 0/1 Embedding 0/1 0/1 0/1 Logistic Regression (LR) All labels L1 L2 L3 …… Multiple Classification (DNN 2-channel inputs) Multiple Binary Classification (DNN 2-channel inputs)
- 18. Intelligent Interaction Technical Practice Clas s2vec Intent rule Intetion model Sen representa tion • Cosine • WMD Intent decision & feature generation Similarity model Data Domain2:top up Intent relation Domain2:weather report Domain1:book airline tickets SM graph Topic model LDA aiml • Lda2vec • skip- thoughts • Maxent • CNN BotFramework B A DC Algorithm component feature generation feature generation feature generation feature generation feature generation feature generation feature generation S2S model RNN • SentenceRNN • ContextRNN
- 19. Intelligent Interaction Technical Practice -QA Bot An Excerpt of a Knowledge Graph Pros Supports Contextual Reasoning Minimized maintaining costs for knowledge items Improved accuracy (10%+) and user experience Cons Loss of recall at initial stage
- 20. Intelligent Interaction Technical Practice - QA Bot
- 21. Intelligent Interaction Technical Practice - QA Bot Preprocessing 1. Anaphora resolution 2. Word segmentation 3. Error correction Recall 1. Open search Answer Processing 1. Answer rendering 2. Logging Computation 1. Similarity 2. Sentiment analysis 3. Attribute identification Indexing 1. Inverted Indexing Structured Knowledge Term Weighting Question Answer Retrieval Model
- 22. Intelligent Interaction Technical Practice -Task Bot Task-Oriented QA Domain Data Input DMS Slots Filling Context Model Output Input：Extract the slots of query Output：DMS feedbacks the result Processes： •Get the attributes of the intent tree to fill •Check the state of the intent tree •Get the result by the state
- 23. Intelligent Interaction Technical Practice -Task Bot AliMe.ai • Features: • Custom chat flow • Custom entities and slot values • Support the 3rd party interfaces to be invoked with specific protocol
- 24. Intelligent Interaction Technical Practice -Task Bot RL for task-bot related-work from MSR References Williams J D, Zweig G. End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning[J]. 2016.
- 25. Intelligent Interaction Technical Practice -Task Bot DRL for task-bot related-work【From Cambrige】 References Wen T H, Vandyke D, Mrksic N, et al. A Network-based End-to-End Trainable Task-oriented Dialogue System[J]. 2016. task generation action generation dialogue managerment belief-tracking intent representation context slot-extraction pattern-extration word seg ner task generation reinforcement learning data preprocess
- 26. Intelligent Interaction Technical Practice - Chit Chat Hybrid Approach: Retrieval Model + Generation • Retrieval Model and Generation Model： • Retireval Model：More platform, but only output answers in a pre-established knowledge base • Generation Model： Generate answers out of the box, but the generated answers sometimes can be inconsistent or meanless • Basic ideas： • Hybrid Approach: IR Rerank + Generation
- 27. Intelligent Interaction Technical Practice - Chit Chat Seq2Seq Rerank Module
- 28. Intelligent Interaction Technical Practice - Chit Chat Experiment
- 29. Intelligent Interaction Technical Practice - Chit Chat Hybrid Approach: IR Rerank + Generation
- 30. Outline AliMe Bot Platform Introduction Intelligent Interaction Technical Practice Look Forward to The Future
- 31. Look Forward to The Future Model construction in different domains and scenes The accumulation of domain data and knowledge is very important The direction of continuous exploration in the field of Technology: generative model, reinforcement learning, transfer learning, machine reading, emotion and so on Construction and application of small data model
- 32. Thank You!