A chatbot developed in Python using Google Maps API
Branch: master
Clone or download
vishakha-lall Merge pull request #1 from janakrajchadha/master
Cleaning imports, removing instances of api key in code, etc.
Latest commit 6021c83 Oct 14, 2018

README.md

MapBot

The project aims to give users a new way to interact with Google Maps by building engaging text-based conversational interfaces. Using Natural Language Processing, it analyses the user's intent and responds in the most useful way.
This project was attempted as a part of the Learn It, Girl mentorship program. The progress through the weeks is as follows:

  1. Week 1: Contructing a roadmap of the project, completing learning material
  2. Week 2: TestBlog example
  3. Week 3: SimpleBot toy chatbot
  4. Week 4: Working with NLP(TestNLTK and StanfordCoreNLP), introduction to GoogleMapsAPI
  5. Week 5,6,7 and 8: MapBotChatBot-tokenisation, grammar parsing, extraction of root word, classification and setting up database connector
  6. Week 9: Integration with Facebook Messenger API
  7. Week 10: Integration of GoogleMapsAPI
  8. Week 11 and 12: Testing and response improvement, updating documentation



TestBlog

A blog example created using Django. The blog provides functionality of viewing blog posts, adding new blog posts and editing existing blog posts. It also includes security features so that changes can be made only by an authenticated user.

SimpleBot

SimpleBot is a toy chatbot developed in Python with a mySQL database backend. The bot stores a table of word associations for responses to a previous sentence and uses this to match future responses. It learns from the previous conversation with the user. It maintains a database of previous replies to the same questions and responds based on queries from the database. It does not extract the meaning of sentences written by the user. When the user types a message, it is understood as an answer to previous statement made by the chatbot. The sentence typed by the user will then be associated with the words present in the previous message. The human message is decomposed in words. The program will try to identify which sentences correspond best to those words, according to its previous “experience”. The limitations of the SimpleBot are the motivation for using NLP and ML.

TestNLTK

Python NLTK
Stanford CoreNLP
Illustraions to understand basic NLP functionalities

  • Exploring corpora and treebanks
  • Tokenisation of text and parts of speech
  • Grammar structure and dependency trees
  • Classification of sentences using ML

TestMapsAPI

Google Maps API
Exploring functionalities

  • Geocoding a location
  • Reverse geocoding
  • Directions between 2 locations
  • Distance matrix between multiple locations
  • Searching for a nearby location

MapBotChatBot

An improvement over SimpleBot to enhance conversation between the user and Mapbot.

  • User input
  • Grammar tokenisation using Standford NLP
    • Extract root word
    • Extract subject word and subject list(for compound entities)
    • Extract object word and object list(for compound entities)
    • Extract all proper nouns(location details)
    • Extract verb
  • Classification of sentence into Statement, Question and Chat entity using feature extractor ML model
  • Store in database depending on the sentence type
  • Respond question with corresponding statement query
  • Respont chat with chat query


Python Library Dependencies

  • numpy 1.13.3
  • pandas 0.21.0
  • scikit-learn 0.19.1
  • googlemaps 2.5.1
  • django 1.11.8
  • urllib3 1.22

Files and Components

MapBotFacebook is the main project folder, it consists of the following components:

  • features.py Feature Generator This Python module extracts features from a sentence. The features.py module includes a function get_triples(pos) which returns a string of the form "POS-POS-POS" where "POS" is a Part-Of-Speech tag.
  • featuresDump.py is a list of extracted features. featuresDump.csv is a dump of features extracted from the sentences.csv using the Feature Generator. This data is then used to train a Random Forrest Model to classify a sentence as a Chat, Statement or Question. The Question and Statement predictions are reported as greater than 80% accurate and the features extraction method could easily be expanded on and enhanced. Also the training data-set is small.
  • googleMapsApiModule.py is a module with several googleMaps functions.
    • direction(origin,destination) Processes the origin and destination request to obtain the directions and display the result in the default browser.
    • add_to_maps_database(origin,destination) Add field to the maps table.
    • get_from_maps_database() Retrive field from the maps table.
    • geocoding(search_location) Processes search location to get geocoded result, and displays the result in the default browser.
  • manage.py
  • requirements.txt
  • utilities.py Multiple functions to perform diverse tasks.
  • chatbot.py Main chatbot module.

Install and Setup

  • Install Python 3.6
  • Install requirements.txt
  • Create an API key for Google Maps API on the Google Developer site
  • Create a new Facebook app on the Facebook Developer site
  • Setup Django server
  • Setup secure tunnels to localhost using ngrok
  • Create a file config.py in the root directory with the following parameter values
  • Send and receive messages

    Check out the Medium article I posted about building this bot.