The article explains the purpose of coarse ranking in industrial search pipelines, outlines key evaluation metrics, discusses sample construction and model architecture choices, and highlights trade‑offs between consistency with downstream ranking and overall system performance.
This article systematically reviews the coarse‑ranking stage of recommendation systems, comparing it with recall and fine‑ranking, defining evaluation metrics, detailing sample design, presenting two technical routes, and exploring optimization directions such as dual‑tower models, knowledge distillation, lightweight fully‑connected layers, multi‑objective and multi‑scenario modeling, followed by practical case studies and results.
The paper presents a two‑stage e‑commerce search framework where the coarse‑ranking stage is redesigned with multi‑objective optimization, expanded negative sampling, and listwise distillation—guided by a new global transaction hitrate metric—enabling it to surpass fine‑ranking on large candidate sets and boost overall GMV by about one percent.
This report details the multi‑stage architecture of Taobao’s main search, introduces a new global‑transaction hitrate metric, analyzes offline and online evaluation gaps, and presents a series of model, loss‑function, and sampling improvements that together lifted overall conversion by about one percent.
Coarse ranking bridges recall and fine ranking by trimming tens of thousands of candidates to a few hundred or thousand using a three‑part framework—sample construction, ordinary and cross‑feature engineering, and evolving deep models—from rule‑based to lightweight MLPs, while employing distillation, feature crossing, pruning, quantization, and bias mitigation to balance accuracy with strict latency constraints.
This article details the business background, system architecture, coarse‑ranking algorithms, dual‑tower model, model distillation, diversity‑control techniques such as DPP, and the online performance gains of the recommendation pipeline used in Tencent Music's Quanmin K‑Song platform.
This article details the practical upgrades of iQIYI's short‑video recommendation coarse‑ranking pipeline, moving from a GBDT model to a dual‑tower DNN, applying knowledge distillation, embedding compression, inference optimizations, and finally a cascade architecture to align with the fine‑ranking model while reducing resource consumption.
iQIYI’s short‑video recommendation team replaced a GBDT coarse‑ranking model with a lightweight dual‑tower DNN, applied knowledge distillation, sparse‑aware embedding optimization, and inference merging, then introduced a cascade MMOE architecture, achieving comparable accuracy with half the memory, ~19 ms latency reduction, and measurable gains in watch time, CTR and engagement.
Meituan‑Dianping’s DSP coarse‑ranking filters large ad candidate sets by scoring ads with user‑profile, weather, and keyword scenario models—using frequent‑itemset mining, AdaBoost, and TF/IDF—then aggregates these scores via a linear‑regression model to select high‑relevance ads for fine‑ranking, boosting click‑through and conversion rates.
