[Paper Review] Comprehensive Movie Recommendation System

Hrisav Bhowmick, Ananda Chatterjee+1 more|arXiv (Cornell University)|Dec 23, 2021

Image and Video Quality Assessment10 citations

TL;DR

This paper presents a comprehensive movie recommendation system integrating content-based filtering (TF-IDF, SVD), collaborative filtering (Pearson correlation, cosine similarity, KNN), and a novel clustering approach using genre-based inertia optimization. Evaluated on the MovieLens dataset (610 users, 9,742 movies), the hybrid framework improves recommendation accuracy by combining multiple algorithms and validating cluster stability via inertia analysis.

ABSTRACT

A recommender system, also known as a recommendation system, is a type of information filtering system that attempts to forecast a user's rating or preference for an item. This article designs and implements a complete movie recommendation system prototype based on the Genre, Pearson Correlation Coefficient, Cosine Similarity, KNN-Based, Content-Based Filtering using TFIDF and SVD, Collaborative Filtering using TFIDF and SVD, Surprise Library based recommendation system technology. Apart from that in this paper, we present a novel idea that applies machine learning techniques to construct a cluster for the movie based on genres and then observes the inertia value number of clusters were defined. The constraints of the approaches discussed in this work have been described, as well as how one strategy overcomes the disadvantages of another. The whole work has been done on the dataset Movie Lens present at the group lens website which contains 100836 ratings and 3683 tag applications across 9742 movies. These data were created by 610 users between March 29, 1996, and September 24, 2018.

Motivation & Objective

• To design and implement a unified movie recommendation system integrating multiple filtering techniques.
• To evaluate the strengths and limitations of individual recommendation approaches (content-based, collaborative, clustering).
• To introduce a novel clustering strategy based on movie genres and inertia to optimize the number of clusters.
• To compare performance across diverse recommendation algorithms using a real-world dataset.
• To provide a comprehensive framework that mitigates the cold start and sparsity problems in recommender systems.

Proposed method

• The system employs content-based filtering using TF-IDF vectorization of movie features and SVD for dimensionality reduction and latent factor modeling.
• Collaborative filtering is implemented via Pearson correlation coefficient and cosine similarity to measure user-item similarity.
• K-Nearest Neighbors (KNN) is applied to predict ratings based on similar users or items.
• A new clustering method groups movies by genre and uses the elbow method with inertia to determine optimal cluster count.
• The Surprise library is used to implement and benchmark collaborative filtering models.
• All methods are evaluated on the MovieLens dataset (100,836 ratings, 3,683 tags, 9,742 movies, 610 users) spanning 1996–2018.

Experimental results

Research questions

• RQ1 How do different recommendation techniques (content-based, collaborative, clustering) compare in accuracy and robustness on the MovieLens dataset?
• RQ2 Can genre-based clustering with inertia optimization improve recommendation performance and cluster stability?
• RQ3 What are the key limitations of individual recommendation strategies, and how can they be mitigated through hybrid integration?
• RQ4 How does the combination of TF-IDF, SVD, and collaborative filtering enhance overall recommendation quality?
• RQ5 What is the optimal number of clusters for movie grouping based on genre and inertia analysis?

Key findings

• The hybrid recommendation system outperforms individual methods by combining the strengths of content-based and collaborative filtering techniques.
• The use of inertia in genre-based clustering helps identify an optimal number of clusters, improving the stability and interpretability of the recommendation groups.
• SVD-based dimensionality reduction effectively reduces noise and improves performance in both content-based and collaborative filtering models.
• Collaborative filtering using Pearson correlation and cosine similarity shows strong performance in rating prediction, especially in dense user-item interaction regions.
• The KNN-based approach demonstrates robustness in handling sparse data by leveraging similar user or item profiles.
• The system successfully addresses cold start and sparsity issues by integrating multiple algorithmic strategies on a real-world dataset.