This is the prototype created in activity 2.2 of the project "AI Assistant for Multilingual Meeting Management" (No. of the Contract/Agreement: 1.1.1.1/19/A/082).
Meeting management is provided using the Virtual Assistant created and edited through the Tilde Bot Dashboard. The .json files included in this site are exported from the Tilde Bot Dashboard and can be imported back to restore the Virtual Assistant.
The Virtual Assistant bot project on the Tilde Bot Dashboard consists of several parts:
- you can define entities that will be recognized in received text (files /entity/en-en.json and /entity/lv-lv.json)
- you can provide utterance examples that are used for intent detector training (files /intent/en-en.json and /intent/lv-lv.json)
- you can define dialog scenario according to which the conversation flow is managed, and appropriate answers prepared (file /model.json)
- you can define actions that are used in scenario to get or process some data, for example, calling Javascript function or the Web Service (file /actions.json)
- you can specify text strings for the output (files /lang/en-en.json and /lang/lv-lv.json)
The general settings of the Virtual Assistant are specified in the file /settings.json
Entities, utterance examples and output text strings in the Virtual Assistant project are prepared in several languages. The language of input is detected automatically, and the appropriate textual data are used for processing of input and preparation of answer.
When the Virtual Assistant receives the input it forwards it to the NLP components - Named Entity Recognition (NER) engine and Intent Recognition module.
The Virtual Assistant recognizes 3 intents:
- intent hello, represented by such utterances as 'hi there', 'hello bot', 'good morning', etc.
- intent bye, represented by such utterances as 'see you later', 'goodbye', 'see you tomorrow', etc.
- intent help, represented by such utterances as 'do you hear me', 'tell me what you can do', 'to what questions you can answer', etc.
The Virtual Assistant recognizes 12 types of entities:
- type Person_name - different first names
- type Speaker - regular expression with two words starting with a capital letter
- type Com_nextmeeting - words 'nextmeeting' or 'meeting'
- type Prompt - regular expression containing words 'Tilde', 'please', and 'write' or 'note' or 'task' or 'decision'
- type Language - regular expression containing name of a language
- type Commands - regular expression containing different commands
- type Com_decision - word 'decision'
- type Com_task - word 'task'
- type Com_stop - word 'stop'
- type Com_download - word 'download'
- type Com_sr_lang - words 'SR language'
- type Com_display_lang - words 'display language'
The Meeting Management Application that works through the Microsoft Teams interface communicates with the Virtual Assistant in order to detect the certain commands and to provide an appropriate answer.
The following commands are implemented:
- command Next meeting
For meeting scheduling the entities of type Prompt and Com_nextmeeting are used. The regular expression captures the text after the initial phrase that contains the time of the next meeting. The answer contains the prefix ‘NEXT MEETING:’ and the passed in utterance.
- command Decision
To use this command user needs to include in the utterance entities of type Prompt and Com_decision. The answer contains the prefix ‘DECISION:’ and the passed in utterance.
- command Task
To use this command user needs to include in the utterance entities of type Prompt and Com_task. The answer contains the prefix ‘TASK:’ and the passed in utterance.
The responses from the Virtual Assistant (tasks, decisions, next meeting times) produced during the meeting are included in the meeting minutes document and are available after the meeting. Meeting minutes document contains also the list of meeting participants and full conversation history.