Cómo construir un chatbot inteligente sin morir en el intento
0 likes114 views
The document discusses the complexities and considerations involved in building intelligent chatbots, including the integration of natural language processing (NLP) and various software engineering practices. It outlines the evolution of chatbot technology, predictions for its market growth, and the tools available for creating effective bots, emphasizing the importance of understanding intents and entities in user interactions. Additionally, it highlights Xatkit as a platform for developing chatbots with a specific domain-specific language (DSL) aimed at simplifying the process.
Cómo construir un chatbot inteligente sin morir en el intento
- 1. 4/4/2022 1 Como construir chatbots inteligentes sin morir en el intento @JordiCabot / @ingdesoftware / @xatkit– jordicabot.com / ingenieriadesoftware.es Introduction 1 2
- 2. 4/4/2022 2 About SOM Research Lab Software runs the world. Models run the software 3 4
- 3. 4/4/2022 3 Nuestra misión Interested in the broad area of systems and software engineering, especially promoting the rigorous use of software models and engineering principles in all software engineering tasks. Flickr/clement127 5 6
- 4. 4/4/2022 4 Context Can you help us create a chatbot to help newcomers to write good bug reports on GitHub? Sure! It’s just a set of questions & answers, this will be pretty simple! Narrator It wasn’t. Once upon a time in 2018 (we were young and naïve) 7 8
- 5. 4/4/2022 5 We were asked to build a bot. We ended up building a bot platform But at least we’ve learnt some things along the way that today we want to share with you 3 times so far! Why you all need to have some bot building notions 9 10
- 6. 4/4/2022 6 And more predictions: - By 2024, consumer retail spend via chatbots worldwide will reach $142 billion—up from just $2.8 billion in 2019. Chatbot market is worth over 1 Billion USD 11 12
- 7. 4/4/2022 7 Chatbots <-> Software engineers • Software Engineers -> Chatbot – Chatbots are a new type of software artefact that we’ll need to consider in our dev projects • Chatbot -> Software Engineering – Bots are more and more playing a supportive role in many SE tasks (e.g. in GH) so better to know how they work and their limitations 13 14
- 8. 4/4/2022 8 Hidden Technical debt in Machine Learning Systems - Google - https://proceedings.neurips.cc/paper/2015/file/86df7dcfd896fcaf2674f757a 2463eba-Paper.pdf 15 16
- 9. 4/4/2022 9 Core chatbot concepts What is exactly a Chatbot? • Bot are software components • Chatbot = Bot + Chat interface – Tb llamado asistente digital, asistente virtual,… • Voicebot = Chatbot + Voice Interface Voicebot Chatbot Bot 17 18
- 10. 4/4/2022 10 VoiceBot • Alexa, Google Assistant ,… • They all hide a chatbot inside 1. Speech-to-text 2. Process the text as a chatbot 3. Text-to-speech to reply Chatbot vs LiveChat • LiveChat: a person is behind • They can be combined – Bot Person: when the bot doesn’t know how to answer (or the visitor is specially “interesting”) – Person Bot: when no support person is available or the client is not worth 19 20
- 13. 4/4/2022 13 Types of chatbots Types of chatbots Button- based (no AI) Intent-based Open questions Open domain 25 26
- 14. 4/4/2022 14 We ignore open-domain bots (e.g. LaMDA) • They can chat about everything but – A complete black-box – ``Useless’’ for a specific business https://mindmatters.ai/2022/03/the-ai-illusion-state-of-the-art-chatbots-arent-what- they-seem/ Creating a chatbot 1 – Why is it so challeging? 27 28
- 15. 4/4/2022 15 Software (bot part) NLU (chat part) Complexity of chatbots Chat bots Many pieces to assemble Input/Output Messaging channels External Platforms NLU Engine Events! 29 30
- 16. 4/4/2022 16 Chatbots are complex software systems Conversation Logic Text Processing External Services Messaging Platforms Deployment Evolution Maintenance Tests J Cabot, L Burgueño, R Clarisó, G Daniel, J Perianez- Pascual, R Rodríguez-Echeverría: Testing challenges for NLP-intensive bots. BotSE@ICSE 2021: 31-34 Creating a chatbot 2 – The NLP component 31 32
- 17. 4/4/2022 17 How NLP works in chatbots Bots and NLP/NLU • At the core of all AI bots there is a NLU/NLP engine with two key missions – Classify the user utterance (i.e. input text) in one of the intents (set of possible questions/requests) the bot understands – Identify the parameters (called « entities ») from the utterance 33 34
- 18. 4/4/2022 18 NLP in a chatbot implies Parameters (entities) I would like to order a medium pepperoni pizza Intent Input text (utterance) • Understand what the user is trying to ask the bot • Identify and gather the data in the request that the bot needs to construct the response 35 36
- 19. 4/4/2022 19 NLU approaches • Regular expressions • Require an almost exact match. Only useful for very concrete cases • Grammar-based approaches • Useful for small domains with a very specific vocabulary and constructs • E.g. Stanford Parser • Neural Networks - Multiclass classifiers 1. You provide some training sentences for each intent (the "classes" ). 2. The classifier learns from these sentences 3. It assigns any new input text to one of the classes (with %confidence) Hi Good morning Good afternoon Hello Morning! Example: bot that greets you, informs about the schedule and lets you buy a pizza Greetings I’d like to buy a pizza I want a pizza Can I order a pizza? Buy Are you open? Is it open now? What are your opening hours? At what time do you start? Schedule If a user says, “I would like to order a pizza now” what is the closest intent? Not that easy when intents start to overlap (e.g. offering a job vs looking for a job: “can I offer you a job?” vs “do you have any job offer”) 37 38
- 20. 4/4/2022 20 Entity recognition • We need to understand the intent but also identify the entities • Much easier if PizzaType is an enumeration ({ Pepperoni, Veggie, Hawaian ,…}) instead of the any string ChooseYourPizza I’d like a PizzaType pizza I’m taking a PizzaType pizza I think I’m going for a PizzaType 39 40
- 21. 4/4/2022 21 Common tricks in NLU Engines • Stemmers identify the lemma of the word (e.g. to focus on the verb, not on the tense of the verb for matching purposes) • Synonyms and paraphrasing are an automatic way to expand the training sentences • Entity augmentation also generalize training sentences • E.g. What’s the weather in Barcelona -> What’s the weather in @city • Augmentation is possible for a large number of data types and common concepts (measures, countries, dates,…) Taking a closer look at the NN https://github.com/xatkit-bot-platform/xatkit-nlu-server 41 42
- 22. 4/4/2022 22 The ML model Even this “simple” arch easily gets over > 100.000 parameters! Step 1 – From words to vectors of numbers 43 44
- 23. 4/4/2022 23 Step 2 Embedding -> similar words have similar vectors King – man + woman = Queen Training and prediction history = model.fit(np.array(context.training_sequences), np.array(context.training_labels), epochs=configuration.num_epochs, verbose=2) full_prediction = context.nlp_model.predict(sequence_from_sentence_to_predict) Tip: Predicting always via the NN is overengineering sometimes (e.g. Yes/No answers) 45 46
- 24. 4/4/2022 24 Remember: the SW part of any ML project And we could go on and on ... 47 48
- 25. 4/4/2022 25 Beyond intent classification • Great bots go beyond providing the right answer, e.g. they could show some empathy • Several language models (for toxicity detection, for sentiment analysis, language identification, emoji support…) can be easily integrated in bots PRO tip: Monitor your bots • You will fail to predict: • What users want to know • How they express their intentions • Careful with self-learning • Filter out what the bot needs to add as training sentences to avoid strange derivations 49 50
- 26. 4/4/2022 26 Creating a chatbot 3 – The Software component What bot building tool should I use? 51 52
- 27. 4/4/2022 27 >100 bot platforms • Rasa, BotPress, Chatfuel, Inbenta, Botsify, Flow XO, Xatkit… • And plenty NLP engines: DialogFlow, Lex, LUIS, Watson, NLP.js It all depends • What are your priorities – Open source? – Free? – Programming language of the bot? – Natural languages (e.g. Catalan) – Availability of connectors – On-premises option? – Plain textual bots or fully functional ones? – Low-code/ no-code or tech skills? – Possibility to migrate from one NLP engine to another? – Monitoring dashboard / reports ? – … 53 54
- 28. 4/4/2022 28 Building bots in Xatkit https://github.com/xatkit-bot-platform Xatkit’s philosophy: Do not reinvent the wheel Other solutions Plenty of great NLU engines and libraries, let’s not waste our time building a new one but take the best of breed and focus on features to help you create better bots Xatkit is a chatbot orchestration platform 55 56
- 29. 4/4/2022 29 Another reason to avoid strong coupling to specific NLU/NLP libraries Xatkit architecture (write your bot once, run it anywhere) Chatbot Designer Intent Recognition NLU Providers (platform-specific) Platform DSL / Libraries Intent DSL Xatkit DSL Chatbot User Instant Messaging Platforms Xatkit Runtime Execution DSL uses uses Platform-independent chatbot definition External Services Deployment Configuration Platform Designer G Daniel , J Cabot , L Deruelle, M Derras : Xatkit: A Multimodal Low-Code Chatbot Development Framework. IEEE Access 8: 15332-15346 (2020) 57 58
- 30. 4/4/2022 30 Xatkit DSL • Bots are created with the Xatkit DSL offering a chatbot- specific syntax for creating: – Intents the bot needs to match – The behavior to execute in response to the matched intents • Our DSL aims to combine the benefits of: – Having a chatbot-specific DSL that facilitates the creation of chatbot applications by providing higher-level chatbot modeling primitives – The full power of a general programming language to create any type of bot, not just simple textual ones Xatkit DSL • Should we go for an external DSL or an internal one? – External one: independent language with its own compiler, editor,… You control everything – Internal ones: lives inside another language. You can reuse the host language but are also constrained by it • We started with an external one and had to reimplement everything In our scenario, Internal DSLs are the way to go (Martin Fowler) 59 60
- 31. 4/4/2022 31 Our DSL is implemented as a Java Fluent API • Create bots using your preferred Java editor – Benefit from all existing Java tooling when developing bots (e.g. debuggers) – Reuse any Java library for complex bot behaviors – Intuitive Fluent Interfaces to help you create advanced conversations – Based on state machine semantics to build any type of bot Intent DSL example 61 62
- 32. 4/4/2022 32 Execution DSL example – State machines Easy deployments A Xatkit bot can be packaged as a single jar with no external dependencies. Just execute the jar to get the bot up and running Via a simple configuration file you can connect your bot with external services, e.g. NLP Engines and communication channels 63 64
- 33. 4/4/2022 33 Xatkit is available on GitHub Advanced topics 65 66
- 34. 4/4/2022 34 Beyond predefined intents 80/20 - Pareto Rule • 80% of user utterances will match 20% of your intents • Recursive Pareto: 64% (80% of 80%) of your input utterances will be reduced to 4% (20% of 20%) of your defined intents. 67 68
- 35. 4/4/2022 35 Open questions -> Plan B in your default fallback https://haystack.deepset.ai/ BODI (Bots for Open Data) we use text to SQL models to try to answer a larger variety of questions No-learn / No-work chatbot building 69 70
- 36. 4/4/2022 36 Low-code / no-code • Many bot platforms try to offer low-code / no- code interfaces to define bots (e.g. GUIs) writing less code • But this is not enough for many non-tech users Keep in mind: 1 language -> ++notations No-Learn (e.g. bots from Excel) 71 72
- 37. 4/4/2022 37 No-Work • Bots derived from your db data – E.g. an eCommerce bot that answers questions about your products, shipping options,… Multiexperience Development Platforms (MXDP) 73 74
- 38. 4/4/2022 38 One app -> Multiple interfaces Related also to Multimodal AI E.g.: DALL-E 75 76
- 39. 4/4/2022 39 E Planas, G Daniel, M Brambilla, J Cabot: Towards a model-driven approach for multiexperience AI-based user interfaces. Softw. Syst. Model. 20(4): 997-1009 (2021) But this brings plenty of new challenges • A chatbot that knows where you’re looking at (e.g. the product in the eCommerce shop) and takes in that info • How do you test that the different UIs behave in a consistent way? 77 78
- 40. 4/4/2022 40 Exciting times ahead The LOCOSS project (Low-code development of smart software) Traditional SW model Smart software model Data Smart back-end AI quality properties Smart front-end Quality dashboard LOCOSS DSLs & Editors LOCOSS infrastructure Explanations Domain Knowledge, Goals & Ethical concerns provides receives Monitoring Testing Explainability specifies Code Generation Local packages Cloud services AI Platforms Training component Code Data Params Deployment component ML Model 79 80
- 41. 4/4/2022 41 To know more: xatkit.com / jordicabot.com / ingenieriadesoftware.es jcabot@uoc.edu @xatkit / @JordiCabot / @ingdesoftware 81