A Factor Marginal Effect Analysis Approach and Its Application in E-Commerce Search System

Algorithm 1

Process of feature importance.

(1)	Input: dataset with features and labels, pretrain model
(2)	Output: Shapley value for the value of the feature
(3)	Required: number of iterations , instance of interest , feature index , data matrix , and machine learning model
(4)	For all :
(5)	Draw random instance from the data matrix
(6)	Choose a random permutation of the feature values
(7)	Order instance x:
(8)	Order instance z:
(9)	Construct two new instances,
(10)	With :
(11)
(12)	Without :
(13)
(14)	Compute marginal contribution:
(15)	Compute the Shapley value as the average:
(16)	End For