DeepLearning Chatbots com RASA e RocketChat

Apresentação feita no FISL18 sobre a experiência de produzir chatbots com DeepLearning sobre a plataforma RASA.AI conectados no Rocket.Chat

Transcript

1. All power of bleeding edge deep learning
   technologies to build clever bots
   & Rasa.ai
   

   View Slide

2. /BOTS
   Mande um e-mail para
   diego.dorgam@rocket.chat
   @diego.dorgam
   ACESSE
   https://softwarelivre.rocket.chat
   

   View Slide

3. Uma comunidade que não para de crescer.
   Uma plataforma de código aberto que permite a
   troca de mensagens em tempo real entre
   colaboradores, clientes e fornecedores relevantes.
   Rocket.Chat STARS
   15.7K
   FORKS
   3.4K
   

   View Slide

4. Em 2015, Gabriel Engel, CEO da
   Rocket.Chat, criou um projeto open
   source na internet para criar a melhor
   plataforma de chats do mundo.
   Com dois anos de projeto, somam-se
   diversos prêmios referente a inovação
   do projeto, um investimento
   americano de 5 milhões de dólares,
   27 colaboradores, mais de 10 milhões
   de usuários e a maior e mais ativa
   comunidade de desenvolvedores na
   categoria de chats no Github.
   O projeto
   

   View Slide

5. > Focus on the product
   > Keep open source
   > Grow the community
   > Experiment on bizmodels
   The Investor >
   

   View Slide

6. Mudança de Paradigma das décadas
   Em cada década vivemos
   um novo paradigma
   tecnológico
   1980's 1990's 2000's 2010's
   

   View Slide

7. Comunicação Corporativa
   Voltada para produtividade
   Desenhada para equipes
   Comunicação interna ágil
   Organiza atendimento externo
   Chatbots Inteligentes
   Bots para automação
   Webhooks de integração
   APIs para interoperabilidade
   Mensagens de Vídeo e Áudio
   A plataforma Startup Category
   OW2 Innovation Award
   1st prize at
   All Things Open
   1st prize at the
   American Chamber of Commerce Arena
   Open Source Rookies
   of the Year 2016
   

   View Slide

8. On-Premises On Cloud
   Implementação
   Instale no seu próprio servidor.
   Ideal para empresas que lidam com
   informações sigilosas.
   Na sua Infraestrutura
   Rode de forma segura e escalável.
   Não se preocupe com a parte
   técnica.
   Na nuvem da RTM
   

   View Slide

9. View Slide

10. View Slide

11. Personalizações
    Change Theme Colors
    

    View Slide

12. Personalizações
    

    View Slide

13. Clientes
    

    View Slide

14. Clientes no Setor Público
    

    View Slide

15. View Slide

16. View Slide

17. Market
    Demand
    The Chatbot explosion
    

    View Slide

18. PACO NATHAN - DIRECTOR O'REILLY MEDIA
    THE AVAILABILITY OF TEMPLATED NLP TOOLS, PLUS THE RISE OF BETTER
    CHAT SYSTEMS, ALONG WITH A NEED TO HANDLE OVERWHELMING
    INBOUND SCALE 24/7, HAVE BOOSTED THE GROWTH OF CHATBOTS.
    

    View Slide

19. SUNDAR PICHAI, GOOGLE CEO
    OVER TIME, THE COMPUTER ITSELF - WHATEVER ITS FORM FACTOR - WILL
    BE AN INTELLIGENT ASSISTANT HELPING YOU THROUGH YOUR DAY. WE
    WILL MOVE FROM MOBILE-FIRST TO AN AI-FIRST WORLD
    

    View Slide

20. An estimative from Business Insider,
    chatbot industry grew up to 229%
    between 2015 and 2016.
    The industry expects to replace the
    equivalent to US$ 23 bi in human
    position salary, just in 2018.
    http://www.businessinsider.com/chatbot-revenue-is-on-
    the-upswing-2017-3
    ChatBots Market growing
    

    View Slide

21. The chatbot magazine conducted a survey on the matter.
    When asked about on which platform companies prefer to
    launch their chatbots, they have chosen Facebook and
    Company Website for the two first positions.
    Those are channels that can be easily reached with
    Rocket.Chat
    https://pt.slideshare.net/Mobileappszen/cha tbots-survey-2017-chatbot-market-resea rch-report
    Chatbot Survey 2017
    

    View Slide

22. About the functions that can be affect by
    chatbots, Customer Services and
    Sales/Marketing automation are the most
    expected to receive significant benefits.
    https://pt.slideshare.net/Mobileappszen/cha tbots-survey-2017-chatbot-market-resea rch-report
    Chatbot Survey 2017
    

    View Slide

23. Most of the companies expect to have
    improvements on the matter of intelligence
    and Conversational Language skills of their
    chatbots.
    https://pt.slideshare.net/Mobileappszen/cha tbots-survey-2017-
    chatbot-market-research-report
    Chatbot Survey 2017
    

    View Slide

24. NEM TODOS OS BOTS
    SÃO IGUAIS
    

    View Slide

25. Process Automation
    Productivity
    Chatbot
    Virtual Assistant
    CATEGORIAS
    Robotic
    Process
    Automation
    Also known as RPA, this is a kind of robot made for
    automate tasks made by humans in a way where the
    robot interacts with a user interface
    This kind of approach is gaining scale at services
    integration where a API is not possible.
    Those are screen scraping robots, that can easily
    assume a human position in tasks like inputting data
    into system forms, collecting data from reports and
    internet, and all kinds of repetitive tasks made by
    humans.
    

    View Slide

26. Process Automation
    Productivity
    Chatbot
    Virtual Assistant
    CATEGORIAS
    Bots for
    Productivity
    Those are bots created to integrate process and
    application into your chat environment.
    By giving a fast approach to productivity systems,
    like git repositories, trello, twitter, or even mass
    internal communication tools, these bots make life
    easier and much more agile.
    One branch of this discussion is the ChatOps
    culture, one of the DevOps abilities that
    organizations must develop.
    

    View Slide

27. Process Automation
    Productivity
    Chatbot
    Virtual Assistant
    CATEGORIAS
    ChatBots
    Most booming technology on the market!
    Chatbots combine natural language understanding
    with service integration technologies (APIs) to
    deliver services and information.
    Chatbots are getting the attention of the big players,
    and offering a new paradigm on how customers
    interact with brands.
    

    View Slide

28. Process Automation
    Productivity
    Chatbot
    Virtual Assistant
    CATEGORIAS
    Virtual Assistants
    Those are very specialized Artificial Intelligence
    bots, that are designed to help humans with some
    personal tasks.
    Organize your schedules, find a restaurant nearby,
    playing a list of your favourite music, those are some
    tasks that Virtual Assistants can do.
    They are intended to help users with the daily tasks,
    by making them easier to access through voice
    commands, for example.
    

    View Slide

29. How chatbots work
    Dialog
    Engine
    NLP API
    Chat
    msg
    Intenção
    1
    Ação 1
    Resposta
    Intenção
    2
    Resposta
    

    View Slide

30. View Slide

31. we wanted more…
    

    View Slide

32. All power of bleeding edge deep learning
    technologies to build clever bots
    Rasa Stack
    

    View Slide

33. STACK
    

    View Slide

34. CORE
    

    View Slide

35. CORE
    domain.yml
    

    View Slide

36. CORE
    domain.yml
    

    View Slide

37. CORE
    domain.yml
    

    View Slide

38. CORE
    domain.yml
    

    View Slide

39. CORE
    nlu.md ## intent:greet
    - hey
    - Hello
    - Hi
    - hello there
    - good morning
    - good evening
    ## intent:goodbye
    - good by
    - cee you later
    - good night
    - good afternoon
    - Bye
    - Goodbye
    ## intent:mood_affirm
    - Yes
    - indeed-
    - of course-
    - that sounds good-
    - Correct
    ## intent:mood_deny
    - No
    - Never
    - I don't think so
    - don't like that
    - no way
    

    View Slide

40. CORE
    stories.md
    ## happy path
    * greet
    - utter_greet
    * mood_great
    - utter_happy
    ## sad path 1
    * greet
    - utter_greet
    * mood_unhappy
    - utter_cheer_up
    - utter_did_that_help
    * mood_affirm
    - utter_happy
    ## sad path 2
    * greet
    - utter_greet
    * mood_unhappy
    - utter_cheer_up
    - utter_did_that_help
    * mood_deny
    - utter_goodbye
    ## say goodbye
    * goodbye
    - utter_goodbye
    

    View Slide

41. CORE
    Interactive
    Learning
    

    View Slide

42. CORE
    nlu_model_config.yml
    pipeline: "spacy_sklearn"
    

    View Slide

43. CORE
    nlu_model_config.yml
    pipeline:
    - name: "nlp_spacy"
    - name: "tokenizer_spacy"
    - name: "intent_featurizer_spacy"
    - name: "intent_classifier_sklearn"
    - name: "ner_crf"
    - name: "ner_synonyms"
    

    View Slide

44. NLU
    RASA Natural Language Understanding
    Um interpretador próprio que pode
    funcionar como um server http:
    "I am looking for Mexican food"
    retorna:
    

    View Slide

45. custom policy
    CORE
    

    View Slide

46. custom policy
    CORE
    

    View Slide

47. GSOC’18 Bot Projects
    

    View Slide

48. GSOC’18 Bot Projects
    

    View Slide

49. GSOC’18 Bot Projects
    OpenSource Bot Frameworks
    Connectors
    https://github.com/
    arthurtemporim
    

    View Slide

50. Hojeexistem mais de 150 mil servidores e
    10 milhões de usuário da Rocket.Chat.
    Em 2018 vamosconectar osservidores
    como se fosse uma redee-mail em tempo
    real, conectando diferentesempresas sem a
    necessidade de criar novosusuários.
    A RedeRocket.Chatfuncionará como uma
    Federação, ondenenhuma empresa detem
    osdados de forma centralizada, e todos
    podem embarcar.
    Futuro da Plataforma
    NO
    GATEKEEPERS
    

    View Slide

51. Estamos trabalhando para disponibilizar um
    grandemarketplace, para que usuários
    compartilhem as aplicações que
    desenvolveram para uso próprio.
    No nosso marketplace será possível
    compartilhar e rentabilizar uma variedade
    de apps, integrações e botsde forma rápida
    e fácil.
    Futuro da Plataforma
    

    View Slide

52. A integração com o BlockChain irá permitir
    que oshistóricosde conversas realizadas no
    Rocket.Chatsejam 100% confiáveis,
    avalisadas por uma redede confiança
    criptográfica, uma das mais inovadoras e
    empolgantes tecnologiasem crescimento
    no cenário atual.
    O setor financeiro éum grande
    impulsionador de desenvolvimento de apps
    usando o Blockchain.
    Futuro da Plataforma
    

    View Slide

53. Dúvidas?
    Experience the next level of team chat
    platform.
    Simple, free, for you.
    Mande um e-mail para
    diego.dorgam@rocket.chat
    @diegodorgam
    Converse com a gente no Rocket.Chat
    Acesse o servidor do time da Rocket.Chat
    no link open.rocket.chat . Teste a
    plataforma e nos encontro nos canais:
    ● #general General discussion
    channel
    ● #support Rocket.Chat support
    channel
    /BOTS
    

    View Slide