Tagged articles
dual-tower model
10 articles · Page 1 of 1
This article presents OPPO's comprehensive advertising recall system, detailing the transition from the old to the new architecture with ANN support, the selection of main‑road recall models, the construction of offline evaluation metrics, sample optimization techniques, model enhancements, multi‑scenario training strategies, and outlook for future improvements.
AdvertisingOffline Evaluationdual-tower model
0 likes · 24 min read

This article presents a comprehensive overview of the background, current multi‑path recall methods, and a series of practical optimizations—including dual‑tower models, enhanced vectors, an unbiased IPW‑based framework, and a travel‑state‑aware deep recall model—applied to Feizhu's homepage recommendation system, with both offline and online experimental results demonstrating click‑through rate improvements.
bias mitigationdual-tower modelrecommendation systems
0 likes · 17 min read

The article outlines a comprehensive knowledge‑base retrieval matching solution—combining PageRank‑enhanced DSL rewriting, keyword and dual‑tower vector recall, contrastive fine‑ranking, and optimized vector‑based ranking—implemented via offline DP training and Sunfish online inference on Milvus, with applications in enterprise search and recommendations and future plans for graph‑neural embeddings.
Contrastive LearningInfoNCEMilvus
0 likes · 12 min read

The Cascade Architecture for Creative Selection (CACS) reorders the advertising pipeline by placing a dual‑tower creative‑selection module ahead of ranking, using soft‑label list‑wise distillation and adaptive dropout to jointly optimize creatives and ads, yielding 5% latency increase but significant CTR and RPM gains in Taobao’s search ads.
ad rankingadaptive dropoutcascade architecture
0 likes · 17 min read

The paper proposes CDAN, a dual‑tower retrieval model that separates item attribute and popularity representations via a Feature Decoupling Module with orthogonal embeddings, aligns head‑tail attribute distributions using MMD and contrastive learning, and jointly trains biased and unbiased towers, achieving higher tail recall, lower exposure concentration, and measurable online click‑through improvements.
Contrastive Learningdomain adaptationdual-tower model
0 likes · 13 min read

This article presents JD's end‑to‑end semantic search recall pipeline, covering multi‑stage recall, a dual‑tower embedding model with multi‑head attention, a heterogeneous graph neural network (SearchGCN), a transformer‑based synonym generation system, and a joint index‑training approach that integrates product quantization to improve recall accuracy and efficiency.
Graph Neural NetworkSemantic Searchdeep learning
0 likes · 17 min read

The talk presents JD's end‑to‑end semantic search recall pipeline, covering multi‑stage retrieval, a dual‑tower embedding model with multi‑head attention, a heterogeneous graph neural network for low‑frequency items, automatic synonym generation via transformer models, and a joint training approach that integrates product quantization directly into the model to improve accuracy and efficiency.
Graph Neural NetworkSemantic Searchdeep learning
0 likes · 16 min read

This article examines the unique properties of novel literature and the difficulties of tag‑based recommendation, then details multi‑modal feature representation, dual‑tower semantic modeling, clustering, and YouTube‑style DNN recall techniques used to improve free novel recommendation systems.
AIdual-tower modelmulti-modal features
0 likes · 9 min read

This article examines the unique properties of novel literature, the difficulties of tag‑based representation, and how multi‑modal AI techniques—including dual‑tower models, feature fusion, clustering, and YouTube‑style DNN recall—are applied to improve recommendation accuracy and user decision‑making.
YouTube DNNdual-tower modelmulti-modal AI
0 likes · 7 min read

The talk outlines how Tencent Advertising leverages deep‑learning visual algorithms—including GCN‑based edge refinement, template generation, AutoML‑driven smart review, and a dual‑tower click‑through‑rate model—to automate creative production, improve ad quality, and enhance user experience across creation, review, and playback stages.
AIAdvertisingAutoML
0 likes · 7 min read
