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.

51 Table 1. Comparative characteristics of algorithmic approaches to recommender systems in e-commerce Criterion Collaborative Filtering (CF) Content-Based Filtering (CBF) Hybrid Models Deep Neural Network Models (DL) Recommendation accuracy (precision, recall, NDCG) High with large- scale data; decreases under data sparsity Moderate; depends on the quality of content features High; stable across different scenarios Very high; particularly for complex patterns Scalability and computational complexity Moderate; requires optimisation High; relies on simple vector similarity computations Moderate–high; depends on the architecture Low–moderate; high computational costs Data requirements Requires large volumes of behavioural data Requires structured content data Requires both behavioural and content data Requires large- scale multimodal datasets Interpretability of results Limited High (explainability through content features) Moderate Low (black-box models) Engineering complexity of integration Moderate; well supported in industrial platforms Low; straightforward integration High; complex architecture Very high; requires MLOps and GPU/TPU infrastructure Source: compiled based on [2; 4; 7] Despite significant progress in the development of algorithmic approaches to recommender systems, their practical deployment in e-commerce is accompanied by a range of engineering challenges related to scalability, architectural integration, and the operational reliability of models (Table 2).