Moviebot Mike

Features

Mike predicts 9 user intents, makes recommendations for about 200 IMDb 7+ rated Bollywood movies from the 2010 - 2019 era based on plot similarity, gives movie plots, and drops movie dialogues in the conversation.

Related Links

• Web app link (Heroku)
• Web app repo (Github)
• Model Training repo (Github)

NOTE: The web app is deployed on a free server which goes into sleep mode and takes about 10s for boot up, once loaded the bot will work as expected.

Data Source

I’m using The Indian Movie Database dataset from Kaggle for the Movie Recommendation part of the project.

Usage Guide

When a user asks for a recommendation, the bot activates the a module that handles such requests, following are commands to trigger that module

• For Movie Recommendations enter “movies like movie-title” command.

  >> movies like rockstar
  

• For Movie Plots enter “about movie-title” command.

  >> about zindagi na milegi dobara
  

• Easter Eggs.

  >> who are you?
  >> what did you do all day?
  

• Casual conversation.

  >> how are you?
  >> do you have friends?
  >> you're mean
  

Architecture

Extended Features

Feature	Description
Unk Handler	Implemented a module to replace out of vocabulary words (words not seen in training) detected in a user request with ‘unk’ to gain maximum information out of the request, the UNK percent in the bot response refers to that.
Movie Title Spellings	To handle variations of movie title spelling, implemented a module that computes Levenstein distance between user input movie title and the ones in the database to capture appropriate movie title given by the user.
Profanity Handler	Implemented an intent that tries to detect swear words in a user’s request.