CCS360 – RECOMMENDER SYSTEMS
NAME: RITHANYA S
CLASS: CSBS B
ASSIGNMENT NO: 1
[Link]: 211722244082
TITLE: Singular Value Decomposition
MARKS AWARDED:/10
FACULTY SIGNATURE
Title: Role of SVD in Enhancing the Performance of Recommender Systems
Abstract
Recommender systems have become an integral part of digital platforms, offering
personalized suggestions to users. One of the key techniques that significantly enhances their
performance is Singular Value Decomposition (SVD). SVD is a powerful matrix
factorization technique used in collaborative filtering, especially to handle sparse data and
latent patterns in user-item interaction matrices. This paper explores how SVD improves the
accuracy and efficiency of recommendations and presents a practical scenario in the context
of a movie recommendation system.
1. Introduction
Recommender systems predict users' preferences for items based on past behavior. In
collaborative filtering, the core component is the user-item matrix, where rows represent
users, columns represent items, and entries denote ratings or interactions.
However, real-world datasets are highly sparse—most users rate only a small number of
items. This is where SVD plays a critical role by reducing the matrix into a set of latent
features, revealing hidden relationships between users and items even when direct
interactions are missing.
2. Understanding SVD
SVD decomposes a matrix RR into three matrices:
R≈UΣVT
Where:
RR is the original user-item matrix.
UU: User-feature matrix.
Σ\Sigma: Diagonal matrix with singular values (importance of features).
VTV^T: Item-feature matrix.
By selecting only, the top-k singular values, we reduce dimensionality and retain only the
most important latent factors, which:
Removes noise
Fills in missing values (matrix completion)
Captures implicit similarities between users and items
3. Benefits of SVD in Recommender Systems
Dimensionality Reduction: Drastically reduces computation time by lowering
feature space.
Noise Filtering: Removes less relevant user-item interactions.
Improved Predictions: Captures patterns that aren't visible in raw data.
Cold Start Mitigation: Predicts preferences even with limited user history using
latent factors.
4. Practical Scenario: Movie Recommendation System
Let’s consider a movie recommendation system like Netflix or Movielens.
The user-item matrix contains ratings (e.g., 1 to 5 stars) given by users to movies.
The matrix is 95% sparse, as most users watch and rate only a few movies.
Using SVD:
o The system identifies latent features (like genre preference, actor affinity,
director style).
o A user who likes "The Dark Knight" and "Inception" (both Christopher Nolan
films) might be recommended "Interstellar" based on latent similarities—even
if the user hasn't rated it.
Example:
Dark Knight Inception Interstellar Titanic
User A 5 4 ? 1
After applying SVD, the system predicts a high score for "Interstellar" for User A based on
their affinity for Nolan's movies and low preference for romance films like "Titanic".
5. Real-World Application
Netflix Prize: SVD was a key technique used by teams participating in the Netflix
Prize competition to improve recommendation accuracy.
Spotify: Uses matrix factorization techniques like SVD to recommend songs based on
listening history.
Amazon: Uses collaborative filtering enhanced by latent factor models to recommend
products.
