COMPARATIVE ANALYSIS OF COLLABORATIVE, CONTENT-BASED AND HYBRID DEEP LEARNING RECOMMENDATION MODELS

Tursynkhan Tursunov, Dinara Kaibassova

2025

Closed access Metadata (CSL-JSON)
DOI 10.53364/24138614_2025_36_1_12
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Recommender systems play a crucial role in personalized content delivery by leveraging user preferences and content attributes. This study evaluates three advanced recommendation models: Neural Collaborative Filtering (NCF), Graph Neural Network-based Content Model (GNN-based Content Model), and Hybrid Neural Network (HNN). Each model integrates deep learning techniques to enhance prediction accuracy and user experience. The NCF model employs a dual-branch structure consisting of Generalized Matrix Factorization (GMF) and a Multi-Layer Perceptron (MLP) to model non-linear user-item interactions. The GNN-based Content Model represents users and items as nodes in a bipartite graph, utilizing Graph Convolutional Networks (GCN) to propagate relational and content-based information across connections. Lastly, the Hybrid Neural Network combines collaborative filtering embeddings with content-based features, aligning content representation within the learned latent space. Our evaluation, based on the MovieLens dataset, demonstrates that the Hybrid Neural Network achieves the highest accuracy (85%), outperforming NCF (80%) and the GNN-based Content Model (77.5%). The hybrid approach benefits from both collaborative and content-driven features, leading to improved user-item match quality. The GNN-based Content Model, despite leveraging structured relationships, struggles with cold-start users due to reliance on content information. These findings suggest that hybrid approaches are more effective in capturing diverse recommendation signals. Future work may focus on integrating transformer-based architectures and reinforcement learning to further enhance recommendation relevance and adaptability.