Personalized movie recommendation system: Tailoring cinematic suggestions

Vivek Krishna, Manish SinghInternational Journal of Applied Research 2018 · DOI

TL;DR

This paper presents a personalized movie recommendation system that leverages a hybrid approach, combining content-based and item-item collaborative filtering. The system aims to provide more accurate and tailored movie suggestions by analyzing movie features and user preferences, outperforming traditional methods.

What Problem Does This Paper Address?

In today's digital age, users are inundated with content, making it challenging to discover movies that align with their tastes. Traditional recommendation systems, while useful, often struggle with issues like data sparsity and scalability. This paper addresses the need for more effective and personalized movie recommendations by proposing a system that combines multiple recommendation strategies to overcome these limitations and enhance the user's movie-watching experience.

Methods

The authors explore several recommendation approaches, including content-based filtering, collaborative filtering (user-based and item-based), and hybrid filtering. For content-based filtering, they utilize movie features like genre, cast, director, and plot summaries, employing techniques like Count Vectorizer and cosine similarity to find similar movies. For collaborative filtering, they focus on item-item similarity, where movies are recommended based on their similarity to items a user has previously liked. The core of their proposed system is a hybrid approach that integrates these methods. They also highlight the potential of deep learning algorithms, such as neural networks and matrix factorization, for improving recommendation accuracy and scalability.

Datasets

The study utilizes two primary datasets:

  • IMDB 5000 Movie Dataset: This dataset contains information on 4806 movies, including descriptions and credits. It was used for developing the content-based recommender system.
  • MovieLens Dataset (ml-latest-small): This dataset includes 600 items, 100,000 ratings, and 3,600 tag applications for 9,000 films. It was used for genre-based item-item collaborative filtering and item-based collaborative filtering.

Key Insights and Ideas

  • Hybrid is Better: The paper strongly suggests that combining different recommendation strategies (hybrid approach) leads to more accurate and effective recommendations than relying on a single method.
  • Deep Learning's Promise: Deep learning models show significant potential for improving recommendation systems by learning complex patterns and relationships, leading to better accuracy and scalability.
  • Content vs. Collaborative Trade-offs: While content-based filtering can be effective, especially for new users or items, collaborative filtering often offers better scalability. The hybrid approach aims to leverage the strengths of both.
  • Feature Engineering is Crucial: The quality of movie features (tags, descriptions, genres) directly impacts the performance of content-based recommendation systems.

Key Results

The authors report that their experiments with deep learning models demonstrated superior performance compared to traditional machine learning algorithms in terms of accuracy and effectiveness. Specifically, hybrid models achieved over 90% accuracy, while CNNs and RNNs achieved over 85%. They also found that content-based filtering performed better than collaborative filtering in terms of accuracy, but collaborative filtering was more scalable. Their implemented genre-based item-item collaborative filtering and item-based collaborative filtering provided good results, with the item-based approach showing more accurate predictions when genre similarity was less distinct.

Limitations and Future Work

Limitations:

  • Genre Limitation: The paper notes that with only 19 distinct genres, genre-based similarity might lead to a large number of recommendations within the same genre, potentially overwhelming the user.
  • Accuracy vs. Scalability: While deep learning models show promise, the paper acknowledges the trade-offs between accuracy and scalability, suggesting that collaborative filtering is more scalable.
  • No Recommendation is 100% Accurate: The authors concede that no recommendation system can achieve perfect accuracy.

Future Work:

  • Hybrid Recommender Systems: Further integration of collaborative filtering, content-based filtering, and user-based approaches for more tailored suggestions.
  • Explainable AI (XAI): Incorporating XAI to provide transparent and interpretable suggestions, boosting user confidence.
  • Social Media Integration: Leveraging social media platforms to gain social context and allow users to share recommendations.
  • Natural Language Processing (NLP): Using NLP to analyze user reviews and comments for deeper sentiment analysis and preference understanding.
  • Contextual Recommendations: Incorporating contextual data like time of day, location, and weather to provide more relevant suggestions.
  • Multi-language Support: Enabling recommendations in multiple languages.
  • Customized Movie Trailers: Generating tailored movie trailers based on user preferences.

Summary updated

Abstract

Recommender schemes are individual of the most profitable and off-course movements of machine education electronics in trade. This is an information draining approach namely used to foretell the inclination of that stoner. The most common fields place the recommender system is used are books, revelation, documents, music, videos, pictures, etc. In this the paper we've projected a film advice system that is to say established a unified filtering approach that form use of the facts given by druggie’s studies them, and further advises the pictures that are best adapted to the stoner at another time. The urged picture list is sifted according to the environments are likely to these pictures by old druggies and uses colourful engine-knowledge algorithms for this purpose. It also helps druggies to find the pictures of their choice established the videotape happening of different druggies in a direct and effective method outside destroying main show up useless scanning. The bestowed recommender scheme creates approvals using colourful types of information and dossier about druggies from the show dataset. The exact can again browse the pieces of advice without difficulty and find a videotape of their choice. Recommender System is a fashion that's used to approve a part or product to a decent established the muted silver in color’s preference’. unified winnowing is an approach that's widely used in recommender orders. Item- article-based unified penetrating is a unified filtering recommender scheme fashion place the mannerly got the approval established the correspondence among the part environments. Then, we present an approach place we calculate the likeness with the particulars established the sort of details. Any item concedes possibility concern further than individual kidney or order. Based on details' inclination to a particular kidney or order we intend a new part- article- based correspondence rhythmical.
BibTeX
@article{krishna2018personalized,
  author = {Vivek Krishna and Manish Singh},
  title = {{Personalized movie recommendation system: Tailoring cinematic suggestions}},
  journal = {International Journal of Applied Research},
  year = {2018},
  doi = {10.22271/allresearch.2018.v4.i11d.11462},
}
Figure