The Use of Machine Learning Algorithms in Recommender Systems: A Systematic Review (1511.05263v4)

Abstract: Recommender systems use algorithms to provide users with product or service recommendations. Recently, these systems have been using machine learning algorithms from the field of artificial intelligence. However, choosing a suitable machine learning algorithm for a recommender system is difficult because of the number of algorithms described in the literature. Researchers and practitioners developing recommender systems are left with little information about the current approaches in algorithm usage. Moreover, the development of a recommender system using a machine learning algorithm often has problems and open questions that must be evaluated, so software engineers know where to focus research efforts. This paper presents a systematic review of the literature that analyzes the use of machine learning algorithms in recommender systems and identifies research opportunities for software engineering research. The study concludes that Bayesian and decision tree algorithms are widely used in recommender systems because of their relative simplicity, and that requirement and design phases of recommender system development appear to offer opportunities for further research.

• The paper identifies Bayesian and Decision Tree methods as widely used, emphasizing their ease of implementation across RS tasks.
• It categorizes ML techniques including neural networks, matrix factorization, and ensemble methods, highlighting trade-offs between accuracy and computational efficiency.
• The review discusses critical software engineering challenges in RS development, noting research gaps in requirements and maintenance phases.

Insights into Machine Learning Algorithms in Recommender Systems: A Systematic Review

The systematic review presented by Portugal, Alencar, and Cowan offers an in-depth exploration of the landscape of ML algorithms employed in recommender systems (RS). As the burgeoning field of recommender systems continues to evolve, this paper provides a structured overview of current ML algorithm applications and highlights the software engineering (SE) challenges associated with their integration. The comprehensive nature of the paper elucidates the complexities involved in RS development and offers a meticulous analysis of research directions and opportunities.

Main Findings

The paper identifies and categorizes frequently utilized ML algorithms in recommender systems. Among these, Bayesian and Decision Tree algorithms emerge as predominant due to their computational simplicity and efficacy in handling various RS tasks. Bayesian methods were referenced in 7 out of the 26 evaluated studies, while Decision Trees followed closely. The authors suggest that these algorithms' relative ease of implementation contributes to their common usage, an insight that bears significance for researchers evaluating algorithmic efficiency versus complexity.

Additional algorithmic approaches such as neural networks, matrix factorization, and gradient descent-based methods were also examined. Neural networks, despite their potential demonstrated in domains like image recognition and autonomous driving, haven’t seen extensive application in RS as of yet, which the authors attribute to the 'black box' nature of neural decision-making processes. Similarly, techniques such as ensemble learning and bandit algorithms are observed, albeit with less prevalence, likely due to their computational demands.

The paper further explores the domain-specific applications of RS, identifying movies, documents, and product reviews as predominant use cases. The accessibility of real-world datasets, such as MovieLens and IMDb for movies, is cited as a pivotal reason for their frequent utilization in research.

Software Engineering Challenges

A core component of this review is its focus on SE challenges in the context of ML-driven RS. The SE lifecycle stages scrutinized include requirements, design, implementation, verification, and maintenance. The results underscore a notable emphasis on the implementation and verification phases, where the majority of reported issues and proposed future work are concentrated. This indicates a predominantly technical focus within the domain and suggests a potential research deficiency in the requirements and maintenance stages.

The systematic review further emphasizes the use of mathematical and statistical methods, such as cosine measures and Pearson correlations, in algorithm implementation. The integration of MapReduce frameworks in some RS projects underscores the influence of Big Data techniques in handling RS computations effectively.

Practical and Theoretical Implications

This review sets the stage for a nuanced understanding of how RS can be optimized through strategic algorithm selection, balancing between complexity and recommendatory precision. The authors' methodological approach to categorizing algorithm usage and identifying SE challenges provides a roadmap for future exploration in SE applications to RS, particularly in underexplored areas such as early-stage requirements gathering and post-deployment maintenance.

Going forward, the research community might focus on developing a cohesive framework for algorithm selection that incorporates both computational efficacy and specific RS application domains. More robust SE practices, particularly in the requirement and maintenance phases, could enhance RS lifecycle management, ensuring sustainable growth and adaptability of RS technologies.

In summary, by synthesizing the interplay of ML algorithms and SE principles in RS development, this paper contributes to a foundational understanding of the current state of ML algorithm integration in RS and opens portals for addressing SE-related challenges in this rapidly advancing field. Future studies might leverage this systematic review to refine RS methodologies and extend the applicability of ML algorithms across diverse recommendation contexts.