Multi-Armed Bandits in Recommendation Systems: A survey of the state-of-the-art and future directions. (1st July 2022)

• Record Type: Journal Article
• Authors: Silva, Nícollas; Werneck, Heitor; Silva, Thiago; Pereira, Adriano C.M.; Rocha, Leonardo
• Is Part Of: Expert systems with applications. Volume 197(2022)
• Publication Date: 2022-07-01
• Subjects: Recommender Systems -- Multi-Armed Bandits -- Systematic literature review -- Recommendation challenges; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.116669
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: Recommender Systems (RSs) have assumed a crucial role in several digital companies by directly affecting their key performance indicators. Nowadays, in this era of big data, the information available about users and items has been continually updated and the application of traditional batch learning paradigms has become more restricted. In this sense, the current efforts in the recommendation field have concerned about this online environment and modeled their systems as a Multi-Armed Bandit (MAB) problem. Nevertheless, there is not a consensus about the best practices to design, perform, and evaluate the MAB implementations in the recommendation field. Thus, this work performs a systematic literature review (SLR) to shed light on this new topic. By inspecting 1327 articles published from the last twenty years (2000–2020), this work: (1) consolidates an updated picture of the main research conducted in this area so far; (2) highlights the most used concepts and methods, their core characteristics, and main limitations; and (3) evaluates the applicability of MAB-based recommendation approaches in some traditional RSs' challenges, such as data sparsity, scalability, cold-start, and explainability. These discussions and analyzes also allow us to identify several gaps in the current literature, providing a strong guideline for future research. Highlights: A literature review of studies about MAB in recommender systems from 2000 to 2020. A discussion about MAB algorithms, datasets, and evaluation metrics. An updated panorama about the current practices and models applied in MAB researches. Discussion on the applicability of MAB models in the main recommendation challenges. Future directions to be explored by using MAB in the recommendation field.