Feature Dual Supervision Model for the Searches of Online Advertising Audiences

Table 1

Notations for the proposed model FDSM.

Symbol	Definition
	The instance contains user and ad characteristics on dataset
	The embedding matrix of instance
	The embedding vector of the th field about instance
/	The characteristics of user ’s and ad ’s
/	User ’s embedding matrix and ad ’s embedding matrix
/	User ’s embedding vector and ad ’s embedding vector
/	User ’s mean pooling vector and ad ’s mean pooling vector
/	Ad ’s supervision vector and user ’s supervision vector
	The output from the last hidden layer in the th fully connected network
/	User ’s multirepresentation and ad ’s multirepresentation from FEU
/	User ’s supervised weight vector and ad ’s supervised weight vector
/	User ’s feature expression vector and ad ’s feature expression vector
	The vector from bridge connection unit
	The prediction of CTR from FDSM