DeepFusion: Fusing User-Generated Content and Item Raw Content towards Personalized Product Recommendation

Table 1

A summary of key notations.

Notation	Definition
	A user
	A item
	The dimension of word embedding
	The length of user review
	User ’s review text consisting of words
	Word vectors of user
	The number of neurons in the convolution layer
	The - kernel in the convolution layer
	The window size of convolution kernel
	The - feature map in the convolutional layer
	The bias of - convolutional kernel
	The output of neuron in the convolutional layer
	The output of the pooling layer
	The weight matrix of the fully connected layer
	The bias of the fully connected layer
	The dense vector of the ID embeddings of user or item
	The dense vector of price range of item
	The price feature of item
	The brand feature of item
	The title feature of item
	The description feature of item
	The concatenate of vector , and
	The latent features of users in the context of reviews
	The latent features of users in the context of ratings
	The latent features of items in the context of reviews
	The latent features of items in the context of ratings
	The latent features of items in the context of item metadata
,	The latent features of users and items, respectively
	The concatenate of vector and
	The predicted value of user on item
	The ground-truth value of user on item
	The dropout ratio
	The number of latent factors
,	Regularization parameters of user and item, respectively
	The learning rate
	The batch size
	A percentage of the length of a review
	A percentage of the number of vocabulary