Word Sequential Using Deep LSTM and Matrix Factorization to Handle Rating Sparse Data for E-Commerce Recommender System

Table 1

Previous state-of-the-art methods.

Method	Description
SVD	A collaborative filtering recommender system based on the latent factor model that is applied by using singular value decomposition (SVD) as the low-rank dimensional factorization, aimed at generating rating prediction [12]
PMF	An advanced version of the SVD model that considers a probabilistic approach to enhance the correspondent users and items. PMF has become a standard rating prediction approach that only involves ratings for collaborative filtering [15]
LDA	An early proposed model that integrates product review document and matrix factorization and aims to interpret the document by exploiting LDA to increase the effectiveness in rating prediction [16]
CTR	A state-of-the-art recommendation model, which combines collaborative filtering (PMF) and topic modeling (LDA) to utilize both ratings and documents [17]
CDL	Another state-of-the-art recommendation model, aimed at enhancing the accuracy of rating prediction by analyzing product documents using a deep learning machine approach based on the autoencoder (AE) that is integrated into the latent factor based on PMF [18]
DCCR	Deep collaborative conjunction recommender (DCCR), a model resulting from multilayer perceptron (MLP) and autoencoder (AE). The autoencoder is responsible for extracting the latent features of an item representation, and the MLP is responsible for detecting the correspondent user and item based on fusion [19]
ConvMF	A collaborative filtering model that involves the traditional matrix factorization model and the document of a product review. Capturing product document understanding involves the convolutional neural network (CNN) with dimensional reduction feature and word embedding [20]. This model is an enhancement of the CDL and CTR models
Att-ConvCF	A version of the collaborative filtering approach, combining matrix factorization and document product review using the attention method in the convolutional process. Matrix factorization is responsible for producing rating prediction [21]
SRCMF	Social review from customer integrated into matrix factorization to achieve effectiveness in generating rating prediction. This approach also requires a product document to be integrated into matrix factorization [22]