Movie Recommendataion System using Content Based Filtering

Introduction

• This project is a movie recommendation system using content based filtering.
• The first dataset which contains information about movies till 2017 used is the TMDB 5000 Movie Dataset which contains information about 5000 movies. The dataset can be found here.
• The dataset for movies from 2018 to 2023 is created using the TMDB API and Wikipedia.
• The names of the movies were scraped from Wikipedia and the data was collected using the TMDB API.
• TMDB API :- TMDB
• Wikipedia links :-
  • 2018
  • 2019
  • 2020
  • 2021
  • 2022
  • 2023

The model

• The model uses the cosine similarity between the movie vectors to find the most similar movies.
• The movie vectors are created using the following features:
  • The genres of the movie
  • The cast of the movie
  • The director of the movie
  • The keywords of the movie
  • The overview of the movie

What is happening in the code ?

1. Stemming:

   • It uses the nltk.stem.porter module's PorterStemmer to perform stemming on the text data in the 'tags' column of the DataFrame (ew_df).
   • Stemming reduces words to their root form (e.g., 'running' to 'run', 'easily' to 'easili'), aiming to normalize the text data for analysis.

2. CountVectorizer:

   • Utilizes CountVectorizer from sklearn.feature_extraction.text to convert the processed 'tags' data into numerical vectors.
   • max_features=5000 sets the maximum number of features to consider.
   • stop_words='english' removes English stop words (common words like 'and', 'the', 'is', etc.) during vectorization.

3. Cosine Similarity:

   • Calculates the cosine similarity between the vectors derived from the 'tags' using cosine_similarity from sklearn.metrics.pairwise.
   - This similarity matrix shows how similar movies are based on their 'tags' content.

4. Recommendation Function (recommend):

   • Finds the index of the given movie in the DataFrame (new_df) based on its title.
   • Retrieves the similarity scores of that movie with all others from the similarity matrix.
   • Sorts the movies by similarity scores and prints the top 5 recommendations (excluding the queried movie).