FAQ-Bot-for-Freshmen

Chat Bot in Action

Contents

Folders

• model folder
  • This folder contains the entire dataset collected by me as a json file as required by rasa-nlu
  • The dataset is a collection of questions asked by students before their admission to IIT Bombay on the Student Mentor Program Page
  • The raw dataset is available here
  • The dataset was prepared using tracy which is GUI for creating json datasets for chatbots
  • It is available in the training_data.json file
  • The remaining files in this folder were created by rasa-nlu engine
• static and templates folder
  • This contains the front end html and css scripts for the chatbot interface. This was created by my mentor Saurabh Kumar.

Main Files

• nlu.py
  • This is the most important part of our chatbot.
  • It contains our Natural Language Understanding Engine which was made using rasa-nlu
  • It uses spacy at its heart. SpaCy is a free open-source library for Natural Language Processing in Python.
  • Given a query sentence, the model preforms multi-class classification into 18 different classes/intents.
• response_dict.py
  • This is used to preform a one-to-one mapping from the classes to a specific reply.
  • We plan to extend this by using small neural networks which will map the intent and entities returned by the NLU to a set of more specific answers
  • We plan to implement this using keras. For this I learnt Keras and implemented a Neural Network on Keras to classify digits from MNIST dataset. The repo can be found here.
  • Due to time constraints, this could not be achieved but I plan to complete this in the future.
• chatbot2.py
  • This file creates a flask server for handling the queries from the frontend.
  • It also starts the nlu, loads the model and makes the backend ready for classifying any query.
  • It uses the nlu.py and response_dict.py files to reply to the queries.
  • This is our older chatbot version.
• chatbot.py
  • This is the latest version of our chatbot.
  • It has an added login interface which allows multiple users and stimultaneous logins in addition to all the features of our older model.