The movie recommendation system plays a crucial role in enhancing user experiences and driving user engagement in the era of extensive content consumption. This project aimed to develop a movie recommendation system using the Item-Based Collaborative Filtering (IBCF) approach. The system leverages user ratings data to identify similarities between movies and generate personalized recommendations for users.
The era of digital entertainment has witnessed an exponential growth in the availability of movies and a vast array of content. However, with such an overwhelming selection, users often struggle to discover movies that align with their personal preferences. This challenge has led to the development of movie recommendation systems, which leverage advanced algorithms to provide tailored suggestions, enhancing the movie-watching experience and facilitating content exploration. In this project, we aim to build a movie recommendation system using item-based collaborative filtering techniques implemented in R.
The primary objective of our project is to create a recommendation engine that suggests relevant movies to users based on their viewing history and preferences. By analyzing the patterns of movie ratings and user-item interactions, our system will identify similarities between movies and users to generate accurate recommendations. Through this, we strive to address the issue of information overload and enable users to discover movies that resonate with their tastes and interests.
The motivation behind this project stems from the growing demand for personalized experiences in the entertainment industry. A robust movie recommendation system has the potential to revolutionize the way users explore and engage with movies, leading to increased customer satisfaction and user retention. Additionally, such systems have the capacity to benefit e-commerce platforms by driving sales, enhancing user engagement, and fostering customer loyalty.
To achieve our objectives, we will utilize the recommenderlab package in R, which provides a comprehensive set of tools for building recommendation systems. By employing item-based collaborative filtering techniques, we will leverage user-item ratings to calculate similarities and generate personalized movie recommendations. Furthermore, we will evaluate the performance of our system using metrics such as accuracy and precision to assess the effectiveness of our recommendations.
In the subsequent sections of this report, we will delve into the data pre-processing steps, implementation of the recommendation model, evaluation of the system's performance, and the insights gained from our analysis. Through this project, we aim to contribute to the field of movie recommendation systems and provide a practical example of building an effective recommendation engine using R.
The dataset used for building our movie recommendation system is the MovieLens Dataset. This dataset comprises of 105339 ratings applied to 10329 movies. It contains information about user ratings, movie IDs, movie titles, genres, and timestamps.
The movie_data dataset provides details about the movies in the dataset. It includes the movieId, which serves as a unique identifier for each movie, the movie title, and the genres associated with each movie. The summary of this dataset shows that the movieId ranges from 1 to 149532, with a mean value of 31924. The dataset consists of 10329 movies, with titles and genres stored as character variables.
The ratings_data dataset contains information about the ratings given by users to different movies. It includes the userId, movieId, rating, and timestamp. The summary of this dataset reveals that the userId ranges from 1 to 668, with a mean value of 364.9. The movieId ranges from 1 to 149532, with a mean value of 13381. The ratings range from 0.5 to 5, with a mean rating of 3.517. The timestamp indicates the time at which the rating was recorded.
These datasets serve as the foundation for our movie recommendation system. By analyzing the ratings and movie information, we aim to generate personalized movie recommendations for users based on their preferences and viewing history.
The MovieLens Dataset was obtained and used as the primary source of data for this project. The dataset consists of 105339 ratings applied to 10329 movies, providing a comprehensive pool of user preferences and movie information.
This is conducted in three steps:
- Selecting useful data
- Normalizing data
- Binarizing the data
Through this we visualised the top users and movies through a heatmap. Then we visualized the distribution of the average ratings per user.
In the case of some users, there can be high ratings or low ratings provided to all of the watched films. This will act as a bias while implementing the model. In order to remove this, we normalized the data. Normalization is a data preparation procedure to standardize the numerical values in a column to a common scale value. This is done in such a way that there is no distortion in the range of values. Normalization transforms the average value of our ratings column to 0. We then plotted a heatmap that portrays our normalized ratings.
In the final step of the data preparation, in this data science project, we binarized the data. Binarizing the data means that we have two discrete values 1 and 0, which will allow the recommendation system to work more efficiently. We defined a matrix that will consist of 1 if the rating is above 3 and otherwise it will be 0.
- The movie_data and ratings_data were loaded into the R environment.
- The data was explored to gain insights into its structure and summary statistics.
- Data cleaning techniques were applied to handle missing values or inconsistencies, ensuring the dataset's integrity.
- To incorporate movie genres into the recommendation system, a one-hot encoding technique was employed.
- The movie_data was split into multiple columns, with each column representing a genre category.
- A binary value of 1 was assigned to a movie if it belonged to that particular genre, and 0 otherwise.
- The ratings_data was transformed into a rating matrix, with users as rows and movies as columns.
- The rating matrix was converted into a sparse matrix, which is a suitable format for collaborative filtering.
- Collaborative filtering, specifically the Item-Based Collaborative Filtering (IBCF) method, was applied.
- Similarity calculations were performed between movies to identify similar items.
- Recommendations were generated based on the similarity between movies and user preferences.
A bar plot was generated that visualizes the total views for the top films, with movie titles displayed on the x-axis and view counts represented by the heights of the bars.
A plot showing the distribution of the average rating per user was generated.
A histogram using the qplot() function was created to visualize the distribution of the column count.
We obtained the heatmap of the specified subset of the ratingMatrix.
A heatmap of the movie similarities was dispayed, where darker colors indicate higher similarity values between movies.
A heatmap plot that visualizes the similarity matrix of the trained recommendation model was generated.
A heatmap visualization of the top users and movies based on the binary ratings in the good_rated_films matrix was generated.
A heatmap of top users and movies was generated depicting the movieRatings matrix.
A heatmap of the normalized ratings was created for the top users.
A similarity matrix that measures the similarity between users based on their ratings was created.
The required libraries, such as 'recommenderlab' and 'ggplot2', were imported into the R environment.
- The movie_data and ratings_data were loaded from the respective CSV files.
- Summary statistics and exploratory data analysis (EDA) were performed to understand the dataset.
- A one-hot encoding technique was applied to represent movie genres as binary variables.
- The genre matrix was created, capturing the genre preferences for each movie.
- The ratings_data was converted into a rating matrix, where users were rows and movies were columns.
- The rating matrix was further transformed into a sparse matrix format suitable for collaborative filtering.
- The recommenderlab package was utilized to build the recommendation model.
- Item-Based Collaborative Filtering (IBCF) was selected as the recommendation method.
- Similarity calculations were performed to identify movies with similar ratings and user preferences.
- Recommendations were generated for users based on their movie preferences.
- The recommendation system was evaluated using appropriate metrics such as precision, recall, and accuracy.
- Cross-validation techniques were applied to assess the performance of the recommendation model.
- https://www.r-project.org/other-docs.html
- https://github.com/Rpita623/Movie-Recommendation-System-using-R_Project
- https://data-flair.training/blogs/data-science-r-movie-recommendation/
- Ayushi Dubey (Enrolment no. 01601192022, AI ML'26, IGDTUW, Email ID: mailto:adayushi232@gmail.com, GitHub Profile: https://github.com/dubeyayushi)
- Azmeen Khatoon (Enrolment no. 01701192022, AI ML'26, IGDTUW, Email ID: mailto:azmeenkhatoon2704@gmail.com, GitHub Profile: https://github.com/Hustler-01)
- Bhumika Gupta (Enrolment no. 01801192022, AI ML'26, IGDTUW, Email ID: mailto:bhumikag0110@gmail.com, GitHub Profile:https://github.com/Bhumikagupta-110 )