Movie Counsel - Your own 🍿 Movie/Series Recommender-System.

App Introduction

This Streamlit based Web app helps you find the right recommendation for your favourite Movies or TV Shows with an inbuilt Movie Review Sentiment Analyzer tool.

1. Movie Recommeder System

• App Features
  • Content + Popularity Based
  • Based on IMDB Movies Dataset
  • Search and select max 5 Movies of your choice and get recommendation for those movies, with complete details like Cover Photo, Plot, Genre, Runtime, Country, Language, Kind, Director, Star-Cast etc.

Data Analysis

i). Source Data

• Source - IMDB Movies Dataset from Kaggle
• Description - This dataset is having the data of 2.5 Million Movies/series listed on the official website of IMDB
• Features
• id - Movie ID
• name - Name of the Movie
• year - Year of movie release
• rating - Rating of the Movie out of 10
• certificate - Movie Certification
• duration - Duration of the Movie in minutes
• genre - Genre of the Movie
• votes - Number of people who voted for the IMDB rating
• gross_income - Gross Income of the Movie in Million
• directors_id - ID of Directors who have worked on the movie

ii). EDA

• Data Pre-processing and Data Cleaning is done on around 2.5M data records.
• Following Python Packages are used for analysis: -
  • EDA - Pandas, Numpy, re, scikit-learn
  • Data Visualization - plotly, seaborn, matplotlib
• Please refer to this notebook for complete detailed analysis, also check out other files in this 📁,all these are part of the Data Pre-processing and Data Cleaning.

iii). Movie Recommender Model

• Python Package Cinemagoer is used for fetching missing details from IMDB based on Movie's IMDB-ID for most of the records in the dataset.
• movie tags are created for each movie by combining plot details, runtime details, year, genre, director, star-cast etc.
• Nltk's Porter Stemmer is used for stemming the words of movie tag. Stemming in NLP is basically the process of reducing a word to its word stem that affixes to suffixes and prefixes or the roots.
• Scikit-learn's TfidifVectorizer (Term Frequency Inverse Document Frequency) is used to transform text into a meaningful representation of numbers which is used to fit machine algorithm for prediction. Basically it calculates how relevant a word in a series or corpus is to a text. The meaning increases proportionally to the number of times in the text a word appears but is compensated by the word frequency in the corpus (data-set).
• Scikit-learn's Cosine_Similarity matrix is used for finding the closet movies(documents) for a given movie (document). Basically it measures the similarity between two vectors or matrices based on their angle rather than distances like Euclidean or Manhattan etc.
• Please refer to this notebook for complete detailed analysis, also check out other files in this 📁,all these are part of the Data Pre-processing and Data Cleaning.

2. Sentiment Analyzer

Sentiment Analyzer for Movie Reviews is a comprehensive tool designed to evaluate the sentiment of movie reviews. This project is an integral part of the Movie Counsel web application, which empowers users to explore and discover movies tailored to their preferences.

Key Features:

• Sentiment Analyzer is implemented as a robust API using the FastAPI framework.
• The API is hosted on the Railway cloud platform, ensuring scalability, reliability, and ease of deployment.
• Sentiment analysis models are trained on a vast dataset comprising approximately 180k movie reviews sourced from IMDB.
• The reviews are scrapped from IMDB for both Hollywood and Bollywood releases from 2019 to September 2023 with help of Beautiful Soup.

i) Data Preprocessing

The heart of any sentiment analysis model is the quality of its training data. Therefore, the dataset undergoes a rigorous preprocessing phase to optimize its quality for analysis.

Data Cleaning and Preprocessing Tasks Include:

• Correcting data formats to ensure uniformity and consistency with help of Pandas.
• Assigning review labels, i.e., classifying reviews as positive or negative based on the accompanying ratings.
• Removing special characters and symbols from the text, facilitating more accurate sentiment analysis.
• Applying word stemming techniques to further enhance the quality of the text data with help of nltk.
• visit the Google Colab notebooks in this 📁 for detailed analysis.

ii) Model Building

Sentiment Analyzer leverages state-of-the-art machine learning algorithms to create an accurate and robust sentiment classification model.

Model Building Highlights:

• Combination of machine learning algorithms, including Logistic Regression, Complement Naive Bayes, and XGBoost, to achieve precise sentiment classification.
• Incorporation of pretrained models such as roBERTa to expedite the training process and enhance overall performance.
• Continuous model evaluation and refinement to ensure the highest level of sentiment analysis accuracy.
• visit the Google Colab notebooks in this 📁 for detailed analysis.

iii) Web API

To make sentiment analysis accessible and user-friendly, Sentiment Analyzer provides a comprehensive web API. Users can interact with the API to gain insights into the sentiment of movie reviews.

Key API Features:

• Accepts HTTP POST requests containing movie reviews as input.

• Returns the probability of both negative and positive sentiments predicted by each model.

• Enables users to integrate sentiment analysis capabilities into their own applications and projects.

• Request

{
  "reviews": "This movie was absolutely fantastic! I loved every moment of it."
}

• Response - model:[negative score, positive score]

{
    "logistic_regression": {[0.15, 0.85]},
    "complement_naive_bayes": {[0.18, 0.82]},
    "xgboost": {[0.13, 0.87]}
}

Web App

• Streamlit is used for building the web app and Stremlit Cloud is used for hosting the web app.

4. Find the demo below

Untitled.mp4