A Telegram bot that understands question and provides answers.

For anyone who wants a quick adaptation of Natural Language Processing (NLP) and Machine Learning (ML) Techniques in chatbot development.

• Question Parsing
• Co-reference
• Smart Answering
• Interactivity
• Testing and Rating
• Provides Accurate One-line Factual Answer (Not my focus for now)

It's written in Python 3.

• Question Parsing

  • This feature aims to parse questions thoughoutly for better answer extraction
  • Implemented here

• Co-reference

  • To co-refer what users have previously said (e.g. "I want to know more about A." "What's his job?", "his" refers to "A's")
  • Uses neuralcoref for now

• Smart Answering

  • Smarter answer for some question types (e.g. return Google Map for location questions)
  • Only LOC questions are entertained for now
  • if question type is LOC and any question entities is location-related, i.e."CITY", "STATE_OR_PROVINCE", "COUNTRY", "ORGANIZATION", "LOCATION"
  • Return Google Map of the entities to users
  • Plan to return wikipedia summary for DESC questions

• Interactivity

  • To interact with users via casual conversation if factual question is not asked
  • Uses chatterbot to converse with users
  • chatterbot matches input with conversation turns to find the closest utterance and return the corresponding reply

• Testing and Rating

  • To store user's rating for improving the system over time
  • Plan to ask users to rate right/wrong for each factual question answered, and rate overall experience with the chatbot after a session

Current implementation is a Telegram bot: @corefbot You can choose your own platform to build the chatbot Nice functions of python-telegram-bot:

• send_location(latitude, longitude): returns Google Map for users
• Users can use \info command to signal the bot that you are asking for factual answers instead of just sharing your loneliness with the bot