Proceedings of the International scientific and practical conference ―Modern Science: Challenges and Perspectives‖ (February 9-11, 2026) / Publisher website: www.naukainfo.com. - London, United Kingdom, 2026. - 121 p.

46 The digitalisation of commercial processes and the growth of e-commerce have led to a significant increase in the volume of data on user behaviour, product characteristics, and transactional processes [1, p. 10879]. Modern e-commerce platforms operate in an environment characterised by high dynamics of user preferences, product assortments, and competitive market conditions, which require adaptive algorithmic models capable of continuous learning and parameter adjustment [2]. Under these conditions, recommendation algorithms acquire a central role as tools for content personalisation, optimisation of customer interaction, and enhancement of sales performance. Contemporary recommendation algorithms in e-commerce are developed at the intersection of machine learning, big data analytics, and software engineering, and perform the function of intelligent personalisation of content, product offerings, and marketing communications [3, p. 49]. In academic literature, recommender systems are classified according to data sources, mathematical models, and architectural characteristics, which makes it possible to systematise approaches to their design and deployment. According to the type of data utilised, recommendation algorithms are classified into collaborative, content-based, hybrid, and neural network models [4]. Collaborative filtering is based on the analysis of user behavioural patterns and the assumption of similarity in preferences between users or items, whereas content- based filtering algorithms rely on product characteristics and user profiles constructed from metadata and textual descriptions. Hybrid models combine multiple approaches in order to improve recommendation accuracy and to mitigate the cold-start problem, while neural network–based algorithms employ deep learning to model complex non- linear relationships in user and item data [2]. From the perspective of mathematical modelling, recommendation algorithms can be classified into statistical, factorisation-based, graph-based, and deep neural network models. Statistical and matrix factorisation approaches, including Singular Value Decomposition (SVD) and Alternating Least Squares (ALS), are applied to identify latent factors in user–item interaction matrices, whereas graph-based and