PersonalAssistant

Custom personal assistant for recognizing crud operations on alarms and reminders. Its output can be used as input for some alarm or calendar api.

Here is an example of using a chatbot:

1. User input
2. Intent, probability
3. User input with replaced slots
4. Recognized slots
5. Chatbot's answer

Prerequisites

• Python 3.7
• Install libraries from requirements.txt

Dataset

Facebook Multilingual Task Oriented Dataset https://github.com/AtmaHou/Task-Oriented-Dialogue-Dataset-Survey

Training

Library used for chatbot training: https://github.com/tflearn/tflearn

Project structure

• dataset/original_data contains original data from dataset
• dataset/formatted_data/intent_config.json file configures which intents and slots will be filtered from dataset
• dataset/formatted_data/train folder contains parsed and formatted original data
  • invalid_data.tsv contains data from original dataset which is not valid in chatbot context
  • api.txt logs recognized slots, which can be used as input for some alarm or calendar api
• chatbot/data folder contains model and training_data, model is trained on training_data and training_data is created from formatted data
• chatbot/code/settings.py file contains project settings

How to use it

• Chatbot can be run in two modes:
  • Accuracy test mode (checks accuracy on given test dataset)
  • Manual test mode (assistant accepts user input and responds to user, additionaly logs recognized slots)
• Before each mode assistant checks if all necessary files are present on given paths. If some of them are missing, they will be recreated from original dataset
• dataset/formatted_data/intent_config.json file also configures possible responses to user input, which slots can be recognized in which intent and required slots = slots which must be present in user input