In the previous post, I had a brief introduction about QnA Maker – a natural language storage service on the cloud platform. However, in order to be able to use chatbot flexibly and with the trend of being more user-friendly, today I introduce more about LUIS. LUIS is an AI service that stores conversations in the cloud and applies custom machine intelligence to text, natural language, and user conversation to predict the total meaning. can, and provide relevant, detailed information that makes the conversational process easier, more natural, and flexible.
How to use LUIS in chatbot
After Azure LUIS is published, the client will send the text to the API (natural language endpoint) and get the results returned in JSON format.
Step 1: Client application sends the user a sentence (sentence) to the LUIS endpoint in the form of an HTTP request.
Step 2: LUIS allows you to create custom models on demand to add intelligence to your application. Linguistic models take unstructured user input text and return JSON-format responses.
Step 3: Client application uses JSON response to make usage decisions according to personal needs.
The LUIS application provides smart information so that the client can make smart choices. LUIS does not provide those options.
The languages and regions LUIS support
LUIS multi-language application, if you want to implement a LUIS application that can run in multiple languages, you can have a few options below:
- If all the languages you want LUIS to have support, you will develop each LUIS application for each language.
- If you need to perform an application that has a language that LUIS does not support, you can use the Translator service (provided by microsoft) to translate speech, text into a supported language to perform the operation. cooperation.
The languages supported by LUIS
(Image taken from microsoft’s doccument)
- Note that in the Chinese language, LUIs want to use simpler Chinese characters than the traditional charset, so it should be noted in the reference of pre-made domain names for pre-made language information.
- Note that in the Japanese language, LUIS will not parse, so Keigo does not distinguish between Japanese and Japanese, so you need to combine different levels during development. For example で ご ざ い ま す is not the same as で す.で す is not the same as だ.
Basic limitations in LUIS
LUIS has some basic limitations as follows:
The first is to limit the model, intent control, entity, and feature in LUIS.
- App name: The default maximum number of characters.
- Applications: 500 applications per Azure Resource.
- Batch testing: 10 datasets, 1000 words (sentences) per dataset.
- Explicit list: 50 per application.
- External entities: unlimited.
- Intents: 500 per app: 499 Custom Intents and No Required Intents.
- List entities: Parent: 50, child: 20,000. Standard name is * up to default characters. Synonym values have no length limit.
- machine-learning entities + roles: composite, simple, entity role: Limited to 100 parent entities or 330 entities, depending on the limit that users access. Sub components can be nested up to 5 levels, with up to 10 per level.
- Model as a feature: The maximum number of models that can be used as a feature for a particular model is 10. The maximum number of phrase lists used as a feature for a particular model is 10 phrase lists.
- Preview – Dynamic list entities: 2 lists about 1000 per request for the endpoint of the query prediction result.
- Patterns: 500 patterns per application. The maximum length of the template is 400 characters.
- Regular expression entities: 20 entities Maximum 500 characters.
- Roles: 300 roles for each application. 10 roles for each entity (entity).
- Utterance: 500 characters.
- Versions: 100 versions for each application.
- Versions name: 128 characters.
The second is the quota limit based on the key type.
- TPS is the number of transactions in 1 second.
Third is the keyboard combination to control the LUIS website.
- Control + E converts between token (tokens) and entities (entities) on the list of utterances (utterances list).
Fourth is a map of the world area in the use of the LUIS website and the LUIS endpoint APIs
- Below is a table of the distribution of LUIS websites for each region.
- More details about the division of regions in the world by domain name can be found at the following link: https://docs.microsoft.com/en-us/azure/cognitive-services/luis/luis-reference- regions
Above is the basic information I learned before starting to develop applications on the LUIS platform.
The article will probably lack a lot of content, hoping for comments from readers.
Thank you very much!
Reference link: https://docs.microsoft.com/en-us/azure/cognitive-services/luis/