Watson Assistant training via web interface on IBM Cloud. In this repository I will put examples of how to create and train a Chatbot using Watson Assistant service. The scope of this tutorial is to create and train a Skill, that, by definition on IBM Cloud, "contains the training data and machine learning logic that enable your chat bot to understand and help customers". After building a Skill, you will be able to create an Assistant: by definition, it "orchestrates the flow of information between a skill and your customers. Add your skill to an assistant, and then deploy the assistant to the media channel of your choice".
In order to use this code, it's necessary to have:
- An IBM Cloud account. It's free of charge, no credit card required (only a valid email address). Possible to have it leveraging your University partnership with IBM (see here for more info: https://www.ibm.com/industries/it/it/educationandresearch/)
- A Watson Assistant Lite Service (free of charge)
In order to perform the Watson Assistant training, the trainer should use the following features, named Intents and Entities, with the related catalogues.
When the user writes something to a general Chatbot, he has in mind a precise intent for each sentence: if the context of the conversation is a restaurant, a sentence "I'm really hungry" could have the intent "I want to order some food". Most of the Chatbots available in the Internet work by keyword: i.e. if a user writes "I'm not really hungry by now" and the keyword is only "hungry", the Chatbot could interpret this as "Place a food order", even if in this case the intent of the user is not to place a food order. With Watson Assistant, powered by IBM Watson Machine Learning algorithms, you can actually train the Chatbot to recognize the real intent of the user by providing a set of typical sentences as a knowledge base on that intent: as the user writes a sentence to the Chatbot, the algorithm performs an analysis of the sentence, understanding it by a correlation of it's knowledge base. With Watson Assistant you can create your own intents with your own knowledge base, either by using the graphic interface or by uploading a set of sentences in a well structured csv file.
Watson Assistant has a set of intents ready to be used, divided by categories.
Entities are used to enhance the conversation capabilities. The best way to understand entities could be the following: let's create an intent for "food order", that understands sentences from "can I place a food order" to "I'm really hungry...". Let's say that a restaurant menu is full of options: so instead of creating one intent for each menu entry, we can create only one intent for a general "food order", and then we can use entities to specify what kind of food we are about to order. So if the user will say "I'd like to order two margherita pizzas", Watson Assistant will identify:
- intent: #food_order
- entity: @number:2
- entity: @pizza:margherita
The system entities are similar to the Intents Content Catalog, as they are a bunch of ready to use entities: number, percentage, currency, date, time, etc.
The Dialog is a set of nodes, activated when one or more conditions occur, that contain the conversation flow. the nodes are evaluated from the top to the bottom.
The welcome node is called when the convresation starts. It promps the user with a "welcome" sentence as he starts the chatbot. This node should be always the first node in the dialog.
The anything_else node is entered when the chatbot doesn't find a correlation between the user's sentence and a intent. This node should be always the last node in the dialog, in order to be entered after Watson has read all other nodes without finding a match.
The basic features explored in the previous paragraphs allowed to create a simple chatbot understanding and answering our questions. Now with a fwe more sophisticated but easy to use features we can turn our chatbot in a real virtual assistant, allowing users to interact with it in a real natural language conversation.
Context variables are indicated with the prefix "$" and are useful to define the context of the conversation. For example, if the user is talking about food ordering, and asks for info about delivery, the Virtual Assistant should not forget that the context of the conversation is food ordering! With context variables you can set and deepen the whole context of the conversation, keeping track of the main topic, the subtopics and the attributes of the conversation.
Slots are the simplest way to collect information from user inputs. Just activate slots, add as many as you need, and store contents into Context Variables. If a prompt question is added, the slot becomes required; elsewhere, is optional.
As in human conversations, a conversation with Watson Assistant can be enhanced with digressions. You can activate them on a node or folder by clicking on the configuration button (the gear). You can choose whether to activate the digressions to the selected node, both ingoing and/or outgoing. You can also choose whether to allow digressions only from / to digressions that force the return to the original node.
Fuzzy Matching (beta) is a feature that enables the automatic correction / interpretation of the Entities, considering typos / keyboard errors / user errors. For example, if an user writes "Pizza maghrerita" (with typo errors), with Fuzzy matching enabled Watson Assistant can understand the Entity as "Pizza margherita". In some cases, Fuzzy Matching needs to be disabled: for example, if the user must insert a name, surname or a banking code, Watson Assistant should not try to interpretate the user input, risking to make errors.
When creating the dialog, in particular when setting up Watson's answers, it's possible to customize the response type. The available response types are:
- Text: Watson answers with a text. You can add variations to the answer by clicking "Enter response variation". The variations are set to "sequential" by default, meaning that the answers are provided in the selected order. By setting the responses to "random", each time the node is entered, a random answer of the list will be produced.
- Option: Watson answers with a list of options. You can record the selected option into a context variable and use it to perform actions.
- Pause: set up a pause to the dialog. You can choose the pause amount in milliseconds.
- Image: Watson answers with an image.
Important: you can mix the responses types by clicking on "add response" button: Watson will answer with all responses in a row.
Please find tutorials on how to train Watson Assistant here: link to come (Italian spoken, but with English interface). Please note: the following videos are made with the previous version of Watson Assistant interface. The main structural difference is that "Workspaces" has been renamed "Skills". All other new features are described in the new videos.
- Riconoscere gli Intenti: https://www.youtube.com/watch?v=imEN7GtX5zw&t=1s
- Gli Intenti generali: https://www.youtube.com/watch?v=isrVffJ1po8
- Le Entità: https://www.youtube.com/watch?v=rNoa4udCGrc
- Raccogliere informazioni con gli Slot: https://www.youtube.com/watch?v=d3tbYvYz_UU
- Le Digressioni 1/2: https://www.youtube.com/watch?v=Qa4Y15XzpRc&t=1s
- Le Digressioni 2/2: https://www.youtube.com/watch?v=okxMSkfbQVM
- Organizzare Intenti ed Entità: https://www.youtube.com/watch?v=KTkZRF2lLPk
- Raccogliere informazioni testuali da input: https://www.youtube.com/watch?v=4HMBBB51iZQ
- Gestire le Variabili di contesto: https://www.youtube.com/watch?v=Z8pujSseWTY
- Integrazione con altri Servizi: https://www.youtube.com/watch?v=TDMdf15wRW8
- Controllo e debug: https://www.youtube.com/watch?v=7mp-R_lMLyc
- More to come...
- More to come...
Linked video tutorials are not official assets.
Please refer to this Github repository to know how to integrate the Virtual Assistant into a web app:
- PHP: https://github.com/lucacrippa88/watson-assistant-php
- Node.js: https://github.com/lucacrippa88/watson-assistant-nodejs
- Node-RED: https://github.com/lucacrippa88/watson-telegram-bot (includes Watson Assistant, Watson Visual Recognition, Watson Text to Speech, Watson Language Translation)
This is not an official asset. It has been created by me and it's not intended for professional use. However, it follows all guidelines you can find in https://console.bluemix.net/docs/services/conversation/ and in https://www.ibm.com/watson/developercloud/assistant/api/v1/. For Watson Services SLAs, please have a look here: https://www-03.ibm.com/software/sla/sladb.nsf/sla/bm-0038-09. Linked video tutorials are not official assets.
This project uses the Apache License Version 2.0 software license. https://github.com/lucacrippa88/watson-assistant-training/blob/master/LICENSE