Unverified50% confidenceFactExact time

SPLADE和DeepImpact是学习稀疏检索的代表，产生稀疏向量但每个词权重由神经网络学习得出，能做词项扩展

1

Sources

50%

Confidence

Long-term

Relevance

8/31/2026

First Seen

Sources

检索系统七次演进：从BM25到生成式检索的完整技术脉络

Related Entities

SPLADEtechnology DeepImpacttechnology

Related Claims

Unverified深度学习模型浅层网络学习的边缘、纹理等低层特征具有较强领域无关性，深层网络编码的高层语义特征更具领域特异性79% similar Unverified深度学习中Dropout可以被解释为一种贝叶斯近似77% similar Unverified深度学习中的深度指网络包含多个隐藏层75% similar Unverified潜空间指深度学习中输入数据经神经网络编码层处理后被映射到的高维连续向量空间，语义相近的概念在该空间中几何距离更接近75% similar Unverified"Deepfake"一词由"Deep Learning"与"Fake"合成而来75% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/828701

API

curl https://kongchang.com/api/v1/knowledge/claims/828701

MCP

get_claim(id=828701)