Abstract

Recommender systems play a crucial role in modern digital platforms by enhancing user engagement across ecommerce, media streaming, and other online services. One of the most commonly used approaches to recommendation is Collaborative Filtering (CF), which analyzes past interactions between users and items to understand preferences and provide relevant recommendations. However, issues, such as data sparsity, the cold start problem, and item synonymy, reduce its effectiveness in practice. Current deep learning techniques address these problems only partially and typically treat them individually. This paper proposes a hybrid deep learning model called Collaborative Neural Content Fusion (ConCF). The model combines Neural Collaborative Filtering (NCF) with Convolutional Neural Networks (CNNs) on item metadata, including titles and genres. It simultaneously learns the latent interaction patterns and semantic content characteristics in an unsupervised system. To provide a fair comparison, four baseline models, namely Autoencoder, NCF, Convolutional Matrix Factorization (ConMF), and supervised ConMF, were re-implemented under a standardized preprocessing pipeline and evaluation protocol. The experiments were conducted using the MovieLens 1M dataset. The results show that ConCF performs better than the baselines with the smallest Root Mean Square Error (RMSE =0.877) and the greatest Recall@5 (53.70%) while the sparsity is 20%. These findings demonstrate that by leveraging both collaborative and semantic cues, ConCF achieves significant advantages, thereby establishing it as a scalable and generalizable framework for next-generation recommender systems.