6. 方法 ?
?? Predic)on
?–
?Search
?Ads
?CTR
?Predic)on
?
Pre-?‐
process
User
?Feature
Features
FM
Models
Rank-?‐based
?
Ensemble
?
Post-?‐
processSVM-?‐perf
BPR
Meta
?Features
LR
Query
?Feature
Ad
?Feature
Combina)on
?
Feature
22. Model ?vs. ?Feature ?
评价指标 ?
模型 ?
SVM
RF
NN
LR
LinearSVM
NN
LR
Linear
RF
Fea_Set_1
Fea_Set_2
* ?Feature决定 ?UpperBound ?
* ?Model决定接近UpperBound的程度 ?
* ?不同问题下Model的表现是不一样的 ?