Research And Implementation Of Recommendation Algorithms For Web User Personalized Demand Scenarios

In recent years,with the rapid development of mobile communication networks,the problem of information overload has increased.While massive data provides convenience to users,it also increases the difficulty for users to accurately and quickly obtain target information.Although search engines have alleviated the problem of "information overload" to a certain extent,when the target users do not have clear needs,they cannot meet the diverse needs of users.The emergence of recommendation systems solves the above problems.The recommendation system can dig out user preferences from the user's historical behavior information,so as to make personalized and accurate recommendations to users.Recommendation systems are now widely used in Taobao,Net Ease Cloud Music,Tencent Video and other fields.The effective use of recommendation algorithms can improve user satisfaction and at the same time bring better benefits to enterprises.However,the problem of data sparseness and cold start restricts the development of recommendation algorithms and arouses widespread concern in academia.With the rise of deep learning technology,the fusion of recommendation algorithms and deep learning technology has become an effective way to solve the above problems.The research of the subject has theoretical and application value.Based on the retrieval,sorting and extraction of existing research results at home and abroad,this paper proposes two recommendation algorithm models Deep Rec and Deep Rec-Att based on deep learning technology for the data sparse problem faced by traditional recommendation algorithms.On this basis,Simulation experiments and comparisons are carried out around the three sub-category data sets of the Amazon review public data set to verify the effectiveness of the algorithm proposed in this paper.The research results are as follows:(1)Propose a deep neural network Deep Rec recommendation algorithm based on fusion modeling of ratings and reviewsIn order to fully mine the potential feature representations of users and items,the algorithm uses a multi-layer perceptron to mine the potential feature representations of users from the rating data.At the same time,it uses a convolutional neural network to mine the potential feature representations of users and items in the comment text potential feature extraction module.The user's potential feature representation is fused,and the learned hidden feature representation is again introduced into the multilayer perceptron to learn the non-linear feature interaction between the user and the item,predict the user's score,and perform quantitative expression.The experimental results show that the Deep Rec algorithm can effectively deal with the insufficient accuracy of the recommendation results caused by the data sparse problem,and improve user satisfaction.(2)Propose an attention mechanism Deep Rec-Att recommendation algorithm based on comment level dimensionIn the process of extracting potential features of review texts in existing models,each review is usually given the same weight and cannot learn key information.The Deep Rec-Att algorithm introduces an attention mechanism and designs a pool based on the attention mechanism.The level of transformation automatically captures the attention score of each comment,constructs a weight matrix,and quantifies the potential feature representation of users and items from the comment dimension.Deep Rec-Att adds a multi-layer perceptron feature extraction module and a pooling layer to better capture the potential feature representation of users and items.The simulation experiment is based on similar data sets.Compared with the standard pooling algorithm Deep Rec+,the performance is recommended to be further optimized.(3)Implementation of the prototype systemOn the basis of algorithm research,based on the application scenarios of Web image and video user recommendation,and relying on the Deep Rec-Att algorithm recommendation mechanism,a prototype system for personalized recommendation of Web image and video is implemented.The system is deployed in accordance with the MVC architecture and adopts Python development technology to realize the humanoriented precise recommendation of the system's human-computer interaction,which meets the diverse needs of users and has application value.