会话式人工智能服务

互相交流信息是人类经验的核心。数据和它的分析,常常在业务设置中驱动这种通信。本次会议的目的是让你了解人工智能,特别是自然语言交互(NLI),自然语言生成与深学习的进步,可以创造新的和令人兴奋的机会,建立基于分析的聊天机器人。
展开查看详情

1.Conversational Artificial Intelligence Ben Tomlinson & Andre Violante, SAS #AISAIS17

2.Agenda • SAS • Architecture • Demo – Building a bot – Running a bot • Q&A #AISAIS17

3. The Analytics Process DATA DISCOVERY DEPLOYMENT Access Integrat Cleanse Explore Analyze Model Govern Embed Monitor e AUTOMATE & ORCHESTRATE Foundation for Broad Range of Creation of Analytic Analytics Capabilities Assets • Access - All sources • Reporting - BI • Centrally Manage • Build – Merge and transform • Explore - Data Visualization • Deploy in database • Profile – Explore your data • Forecast – Time Series • Create Efficiencies • Fix Quality - Standardize • Statistics – Explain • Control & Adjust • Governance and Lineage • Prediction –ML and AI • Optimize – Best outcome #AISAIS17

4. Analytics Personas 2 3 4 5 1 WHAT DO THEY DO? WHAT DO THEY DO? WHAT DO THEY DO? WHAT DO THEY DO? WHAT DO THEY DO? Performs complex exploratory Understands the business Extends the business analysis, descriptive context of underlying data & is analyst work pattern with Facilitates data preparation, Develops new applications focused on answering a defined segmentation and predictive model deployment and with embedded analytics. data preparation, exploration question. modeling. monitoring. and visualization. Leverages documented, Provides guidance and support Investigate advanced analytic Integrates analytics into public REST APIs to Explores the use of to stakeholders by building techniques and machine existing production systems reference shared models advanced analytic modeling learning. reports and interpreting results. techniques to solve a and business processes. and individual actions. particular business problem Identifies new data needed for using wizards and prebuilt model development and defines logic. analytical tables. WHAT DO THEY NEED? WHAT DO THEY NEED? WHAT DO THEY NEED? WHAT DO THEY NEED? WHAT DO THEY NEED? Flexible, easy-to-use graphical Flexible, easy-to-use graphical A choice of tools. Uptime and reliability. Portability and user interfaces user interfaces. Advanced coding capability. consistency. Some coding capability. Ability to scale methods Agility and consistency. without redefining code. #AISAIS17

5. Diversity Support all users Consistent Visual Interfaces Programming Interfaces API Interfaces #AISAIS17

6.DATA TIME SERIES STATISTICS DISCOVERY FORECASTING DATA & TEXT MACHINE OPERATIONS MINING LEARNING RESEARCH #AISAIS17

7. Natural Language Processing Natural language processing (NLP) is a branch of artificial intelligence that helps computers interpret, analyze, and manipulate human language Natural Language Processing Text Analytics Natural Speech/Audio Language Processing Information Extraction Understanding Text Classification Sentiment Analysis Topic Modeling Natural Machine Summarization Language Translation, … Generation OCR, etc. SAS Visual Text Analytics Natural Language (VTA) Interaction #AISAIS17

8. Chatbot Business Benefits Ø Operational Efficiency Reduce mundane tasks, improve speed and consistency of case handling, lower- cost channel Ø Availability 24/7 always-on access, multiple concurrent sessions Ø Global Reach Bots can be trained to communicate in multiple languages Ø Smarter Responses More opportunities to embed advanced analytics for better outcomes (retention, cross-sell, upsell) #AISAIS17

9. Sample NLU/NLG Use Cases Voice activated/ Automated Call Steering enabled Image or Text and IVR devices and Captioning apps Automated Machine reports, Learning Smart Search summaries, Model or narratives Explanability #AISAIS17

10. Today’s Focus: Chatbot Use Case Voice activated/ Automated Call Steering and enabled devices Image or Text IVR and apps Captioning •Customer service bot •Order fulfillment •Q & A systems Chatbots •Therapy bot (“Woebot”) • Law bot Automated • Health bot Machine reports, Smart Search Learning Model summaries, or Explanability narratives #AISAIS17

11.#AISAIS17

12. Conceptual Architecture If Speech-to-Text NATURAL and Text-to-Speech LANGUAGE are relevant, that is UNDERSTANDING handled by the NLU returns application layer Submit retained predicted intent + new info to and NLU parameters/ slots Submit inputs User input Enough to execution info to YES engine fulfill intent? Response BOT LOGIC NO ACTIONS USER APPLICATION Incomplete intent: prompt for more info Submit outputs to speech NATURAL LANGUAGE template GENERATION #AISAIS17 #AISAIS17

13.Demo