| With the rapid increase in the amount of online information,personalized recommendations have gradually become an effective strategy to overcome information overload.Although existing recommendation systems have been very successful and have produced good recommendations,but they still face challenges such as sparsity,poor interpretability,and the difficulty of capturing the short-term user interests given several items the user just accessed.In recent years,deep neural networks have attracted great attention in academia and industry,and have been widely used in fields such as natural language processing and computer vision.At the same time,studies have shown that applying deep learning to recommendations can improve the quality of recommendations.Among many deep learning-based recommendations,few literatures use recurrent neural networks to model sequence data to improve the quality of recommendations.This paper analyzes and studies the recommendation system based on recurrent neural network,and proposes a new model,as follows:First,a large amount of information exists in reviews written by users in many e-commerce websites.Most recommendation systems exploit the information of reviews,but only evaluate the importance of reviews from the word-level rather than the review-level.If only considering important words in the reviews and ignoring really useful reviews,the performance of the recommendation model will be reduced.Based on this,this paper proposes a joint deep recommendation model based on double-layer attention mechanism(DLALSTM).First,DLALSTM uses bidirectional long short term memory network(BiLSTM)to jointly model the user and item reviews from both word and review levels,and aggregates the review representation and the user or item representation by a double-layer attention mechanism.Then,to make rating prediction,the latent representation of users and items learned from the reviews are incorporated into the user and item latent vectors obtained from the rating matrix.To compare DLALSTM with the commonly used recommendation methods,we conduct experimental evaluation on different domain datasets of Yelp and Amazon.It is found that the performance of DLALSTM exceeds the state of art recommended methods.Meanwhile,the model proposed in this paper can alleviate the sparsity problem to some extent and has good interpretability.Second,the chronological order of user and item interaction history is a key feature in many recommender systems,because the items that users will interact may largely depend on those items that users just accessed recently.However,with the large increase in the number of users and items,sequential recommender systems still face several challenging problems: First,the hardness of modeling the long-term user interests from sparse implicit feedback;Second,the difficulty of capturing the short-term user interests given several items the user just accessed.To address these challenges,this paper proposes a recurrent neural network based sequential recommendation model(LSTMASR),to capture users' long-term and short-term interests.The model consists of a recurrent network attention module and an item-item relationship module,which is integrated with a matrix factorization model and optimized by Bayesian Personalized Ranking(BPR).Specifically,on the one hand,the model models the historical items that the user interacts with through the LSTM's recurrent structure,and then apply the attention mechanism on the output of the LSTM to extract the user's short-term interest preferences.Applying the attention mechanism can effectively assign appropriate weights to historical items that the user interacts with,and extract item information that is important to the user.On the other hand,the model models the relationship between the user-interacted items and the target item,thus providing important auxiliary information to predict the user's sequence behavior,because users may interact with closely related items before.Finally,the comprehensive experimental analysis results of the sequential recommendation method proposed in this paper on three different datasets show that the performance of the recommendation method is better than other commonly used sequential recommendation methods. |
