A Survey of Latent Factor Models in Recommender Systems

TL;DR

The survey addresses the challenge of delivering accurate, scalable recommendations in the presence of sparse, heterogeneous data by organizing latent factor models around learning data, model architectures, learning strategies, and optimization. It systematically catalogs a spectrum of models—from probabilistic and kernelized approaches to graph neural networks—and learning strategies including self-supervised and transfer learning, highlighting how each contributes to handling implicit feedback, trust, and content data. Key contributions include a structured taxonomy across data modalities and model families, a comparative view of performance trends, and identified gaps such as explainability and scalability in neural and graph-based methods. The work emphasizes practical implications for practitioners seeking robust, context-aware recommender systems and outlines avenues for future research in data augmentation, negative sampling, and domain adaptation. Overall, the survey provides a comprehensive, unified perspective on latent-factor modeling in recommender systems and its evolving methodological landscape.

Abstract

Recommender systems are essential tools in the digital era, providing personalized content to users in areas like e-commerce, entertainment, and social media. Among the many approaches developed to create these systems, latent factor models have proven particularly effective. This survey systematically reviews latent factor models in recommender systems, focusing on their core principles, methodologies, and recent advancements. The literature is examined through a structured framework covering learning data, model architecture, learning strategies, and optimization techniques. The analysis includes a taxonomy of contributions and detailed discussions on the types of learning data used, such as implicit feedback, trust, and content data, various models such as probabilistic, nonlinear, and neural models, and an exploration of diverse learning strategies like online learning, transfer learning, and active learning. Furthermore, the survey addresses the optimization strategies used to train latent factor models, improving their performance and scalability. By identifying trends, gaps, and potential research directions, this survey aims to provide valuable insights for researchers and practitioners looking to advance the field of recommender systems.

Figures (10)

• Figure 1: The structure of a latent factor model based recommender system.
• Figure 2: Overview of the structure of this survey.
• Figure 3: An example of using SVD to predict ratings. Initially, the rating matrix $R$ undergoes imputation, where missing ratings are replaced with the average rating of each item. Subsequently, $R$ is decomposed using SVD. The two factors corresponding to the largest singular values are retained to reconstruct an approximation $\tilde{R}$ of $R$.
• Figure 4: An example of using gradient descent to fit the matrix factorization model without regularization to data. Left: Starting from randomly generated values, gradient descent iteratively updates user and item factors until convergence. The final factors are used to predict the missing ratings. Right: The upper plot shows the evolution of the objective function (SSE) and the norm of the gradient throughout the optimization procedure. The bottom plot shows the evolution of user and item factors from their random initial positions to their final ones.
• Figure 5: The architecture of the NCF model he2017neural.