Recommender systems have evolved rapidly to address the challenges of web-scale personalization. This survey provides an exhaustive review of largescale recommendation techniques published recently between 2019 and 2024, covering recent algorithmic advances and practical system-level optimizations. We review methods based on graph neural networks, transformer-based and deep sequence models, social network–aware techniques, multi-interest user modeling, and multi-objective learning. In addition, we survey benchmark datasets— from public collections such as MovieLens, Amazon Reviews, Yelp, and Criteo Click Logs to emerging multi-modal datasets like MicroLens—and discuss their relevance for algorithm development. We also detail infrastructure considerations, including distributed training, deployment architectures (cloud, on-premise, edge), and optimization techniques such as pruning, quantization, and approximate nearest neighbor search. The role of standardized benchmark datasets and appropriate evaluation metrics is also discussed. Insights from leading platforms including Facebook, TikTok, Instagram, Pinterest, WeChat, LinkedIn, Alibaba, and Amazon are integrated to highlight how state-of-the-art research is translated into production systems. This review aims to serve as a comprehensive reference for researchers and practitioners seeking to design recommender systems that are not only accurate but also scalable, efficient, and adaptable to evolving user needs.