Unverified50% confidenceFactExact time
Two-tower retrieval is the mainstream recall architecture for handling large-scale candidate sets in industrial recommender systems, using two independent neural networks to encode users and items respectively into low-dimensional dense vectors, with match quality measured by vector inner product or cosine similarity.
1
Sources
50%
Confidence
Long-term
Relevance
9/21/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe current mainstream industry architecture for recruitment recommendation systems adopts a two-tower model, where one tower encodes the job seeker's profile and the other encodes job information, and the output vectors of both towers are matched via similarity calculation in a shared embedding space71% similarUnverifiedThe two-stage retrieval strategy first uses fast vector retrieval to recall Top-N candidates, then uses a cross-encoder to precisely score those N candidates and select the final Top-K.68% similarUnverifiedThe reranking module typically uses a cross-encoder model to compute relevance scores between queries and documents, performing a secondary ranking of the candidate documents from the initial retrieval.68% similarUnverifiedModern recommendation systems typically adopt a multi-stage architecture: a recall layer filters thousands of candidates from a massive pool, a ranking layer performs fine-grained scoring and sorting, and a strategy layer applies interventions such as diversity control and safety filtering.67% similarUnverified2015年Koch等人提出的孪生网络（Siamese Network）通过学习样本间的相似度度量来实现少样本分类67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/937969API
curl https://kongchang.com/api/v1/knowledge/claims/937969MCP
get_claim(id=937969)