Advancements in recommender systems: a comprehensive analysis based on data, algorithms, and evaluation

Systematic review and analysis of recommender systems (RSs) in emerging technologies, new scenarios, and diverse user needs are essential for understanding their development, strengthening research, and ensuring sustainability. Using 286 research papers from major databases, this study adopts a systematic review approach to summarize current challenges and future directions in RSs related to data, algorithms, and evaluation. Five key research topics emerge: algorithmic improvement, domain applications, user behavior & cognition, data processing & modeling, and social impact & ethics. Collaborative filtering and hybrid techniques dominate, but RS performance is constrained by eight data issues, twelve algorithmic issues, and two evaluation issues. Major challenges include cold start, data sparsity, data poisoning, interest drift, device-cloud collaboration, non-causal driven models, multitask conflicts, offline data leakage, and multi-objective balancing. Potential solutions include integrating physiological signals for multimodal modeling, mitigating data poisoning via user behavior analysis, evaluating generative recommendations through social experiments, fine-tuning pre-trained models for device-cloud resource allocation, enhancing causal inference with deep reinforcement learning, training multi-task models using probability distributions, implementing cross-temporal dataset partitioning, and evaluating RS objectives across the full lifecycle. The reviewed literature is sourced from major international databases, with future research aiming for broader exploration.