Research On Sequential Recommendation Algorithm Based On Deep Learning

In the era of big data and rapid development of information technology,the problem of information overload has become more and more serious,and getting effective information quickly and accurately has become a new challenge.The information overload problem can be solved by recommendation system,which is also a very effective method of information filtering,and it can help users to get information quickly according to their interests and preferences.Sequential recommendation treats user-item interactions as dynamic sequences,and by modeling the sequence of user-item interactions,the next behavior of the user can be predicted based on his or her changing interests.The existing research methods of sequence recommendation have the following two problems: firstly,they have limited ability to model sequences and cannot model long-range sequence dependencies well,and secondly,users’ interests in behavior sequences are complex and variable,and the ability to capture user interest features is insufficient.In this thesis,we conducted an in-depth study to address the above problems and proposed two sequence recommendation models,and the main related works of this thesis are as follows:1.To address the problem that the existing methods have limited ability to model sequences and cannot model long-distance sequence dependencies well,this thesis proposes a sequence recommendation model(GRU-MAMSR)based on gated recurrent unit and multi-head attention mechanism,which has powerful feature abstraction and model expression ability.The problem of long-range sequential dependencies exists in the sequence modeling process,and GRU-MAMSR can improve such a problem.This model can alleviate the gradient disappearance problem and reduce the model parameters during the neural network training process,so that the model can be trained better.2.In order to address the problem that the interests of users in behavioral sequences are complex and variable,and the existing sequential recommendation research methods are not capable of capturing the characteristics of users’ interests,this thesis proposes a sequential recommendation model(CNN-MAMSR)based on convolutional neural network and multi-head attention mechanism,which can effectively understand the preferences of users’ behaviors,better consider the dynamic interests of users,and improve the accuracy of the recommendation model.In addition,CNN-MAMSR can obtain the dependency relationship between different items and better capture users’ interest in items,so the obtained recommendation effect is also better.The sequential recommendation model designed in this thesis is experimented with other better recommendation models on three public datasets Amazon＿Beauty,Movie Lens-1M and Amazon＿Games,and it can be concluded from the experimental results that the model proposed in this thesis can achieve better recommendation results on HR@K,MRR@K and NDCG@K,thus verifying that the GRU-MAMSR and CNN-MAMSR proposed in this thesis are effective.