1 views collected around this technical thread.
This article explores how to build a scalable, high‑performance leaderboard for hundreds of millions of users by comparing traditional database ORDER BY approaches with Redis sorted sets, addressing challenges such as hot keys, memory pressure, persistence risks, and presenting a divide‑and‑conquer implementation strategy.
In 2024 the advertising team deployed an end‑to‑end AIGC pipeline that automatically creates high‑quality ad images, uses the multimodal Reliable Feedback Network and the million‑size RF1M dataset to filter outputs, builds rich offline and online multimodal representations with contrastive and list‑wise learning, and optimizes ranking architecture to deliver scalable, personalized creative selection.
Tech Insight presents concise overviews of ten JD retail technology papers accepted at top AI conferences in 2024, covering topics such as open‑vocabulary object detection, multi‑scenario ranking, diversity‑aware re‑ranking, a diversified product search dataset, semi‑supervised query classification, plug‑in CTR models, and methods to mitigate LLM hallucinations.
This article describes the definition, architecture, and practical applications of explainability in JD's retail recommendation system, covering ranking, model, and traffic explainability, system challenges, data infrastructure, and specific techniques such as SHAP and Integrated Gradients for interpreting model decisions.
This article introduces Retrieval‑Augmented Generation (RAG) as a solution to the hallucination, freshness, and data‑privacy issues of large language models, details its modular architecture, explains the layered system design and hybrid retrieval pipeline, and shares the practical challenges and engineering tricks encountered when deploying RAG in enterprise office scenarios.
This article summarizes Meituan's content search system, detailing the challenges of heterogeneous, high‑frequency local content, the multi‑modal tagging and representation pipeline, recall and ranking optimizations, satisfaction metrics, multi‑objective fusion, heterogeneous mixing, and future directions for improving user experience in local life services.
This report analyzes performance bottlenecks caused by serialization in a search‑recommendation system, presents detailed measurements of request latency, evaluates multiple optimization strategies—including Redis caching, lazy metric handling, and custom byte‑array serialization—and documents the resulting latency reductions and implementation considerations.
This article presents JD.com’s multimodal recommendation system architecture, covering content understanding, multimodal ranking and recall models, practical deployment pipelines, and future research directions such as large‑model integration and supply‑side generation, all illustrated with detailed diagrams and Q&A.
This article presents JD's multimodal content‑understanding framework, detailing its five‑M business characteristics, the architecture of multimodal recall and ranking models, the GMF and MIN modules for semantic alignment and personalization, and future research directions involving large language models and end‑to‑end multimodal encoding.
In real‑time bidding advertising, a multi‑path recall framework quickly filters billions of ads using parallel non‑personalized and personalized strategies—such as hot‑item rules, collaborative‑filtering, skip‑gram vectors, and GraphSAGE embeddings—while respecting targeting constraints, before a ranking stage optimizes eCPM, with effectiveness measured offline and online and future extensions planned with large language models.
This article presents ten crucial intermediate‑to‑advanced SQL concepts—including CTEs, recursive queries, temporary functions, CASE pivots, EXCEPT vs NOT IN, self‑joins, ranking functions, delta calculations, cumulative totals, and date‑time manipulation—each explained with clear examples and code snippets.
The article details a comprehensive architecture for high‑concurrency services—including a Redis‑backed ranking system, distributed locking, Bloom‑filter nickname deduplication, Netty‑based reliable messaging, IoT MQTT/REST integration, live‑streaming pipelines, layered performance tuning, and automated traffic‑replay testing to ensure scalability and robustness.
The advertising engine uses a five‑stage funnel—retrieval, recall, coarse ranking, fine ranking, and re‑ranking—each optimized with specialized indexes, multi‑channel recall, multi‑objective twin‑tower models, deep CTR/CVR predictors, and cold‑start paths, delivering up to 33 % spend growth, 6 % eCPM lift and lower latency while maintaining diversity.
This article reviews the evolution of JD’s recommendation advertising ranking mechanism, covering its economic auction origins, challenges of multi‑material valuation, user interest uncertainty, and multi‑item auction fairness, and describes AI‑driven solutions such as deep auction models and reinforcement‑learning‑based ListVCG.
The article analyzes JD.com's recommendation advertising ranking auction mechanism, detailing its objectives, challenges in traffic value estimation, user interest exploration, and multi‑item auction fairness, and describing the technical evolution from traditional auctions to deep‑learning‑driven solutions.
This article investigates how large language models can be applied to recommendation tasks, describing two usage strategies, various ranking approaches, experimental evaluations on multiple datasets, comparisons with traditional models, and analyses of prompt design, cost, and cold‑start capabilities.
This article explains the fundamental concepts of search engine technology—including forward and inverted indexes, tokenizers, stop words, synonym handling, ranking algorithms, and NLP integration—and provides a concise Go implementation with code examples and performance testing.
The article details how Autohome tackled the cold‑start problem for newly generated content by redesigning the recommendation pipeline, introducing multi‑path recall, refining ranking and re‑ranking formulas, and applying strategic controls, resulting in a rise of cold‑start success rate from 27% to over 99% and a CTR increase from 5% to 14%.
This article investigates how large language models can be applied to recommendation tasks, presenting two usage strategies, experimental evaluations on multiple datasets, comparisons with traditional baselines, and analyses of prompting methods, cost, and cold‑start performance.
This article presents a comprehensive design for Zhuanzhuan's home‑page recommendation pipeline, detailing the system architecture, challenges of traffic efficiency and diversity, and a two‑stage solution that applies Proximal Policy Optimization reinforcement learning in the re‑ranking module and Determinantal Point Process optimization in the coarse‑ranking and traffic‑pool stages, followed by offline simulation, online deployment, and evaluation metrics.