Hybrid Recommendation Algorithm Based On Rating Matrix And Review Text

The rapid development of the Internet and information technology has caused the amount of information on the Web to grow exponentially.The huge information source has caused people to face severe information overload problems.Therefore,a recommendation system that can effectively deal with information overload problems has emerged and has been applied to each field.The most commonly used collaborative filtering algorithm based on rating data in the recommendation system has been widely researched and applied,but the problem of data sparsity has severely restricted the recommendation effect of the model.On the other hand,it only considers the user's rating information on the product and ignores it.A large number of user review information that can be used contains richer and valuable information resources,and a model can be constructed to mine user and product features in the review text for use in the recommendation system.Therefore,in order to alleviate the sparseness of the scoring matrix and improve the quality of recommendations,recommendation models based on multi-source heterogeneous data(comment texts,social networks,etc.)have been proposed one after another,but they often cannot fully exploit the high-level abstract features of users and products.Learning techniques perform well in mining higher-order features.Therefore,for the above problems,it is of great significance to design and implement a hybrid recommendation algorithm by using deep learning technology to process the comment text information and combining with the traditional recommendation algorithm.This article starts with the user-commodity review text,combined with the convolutional neural network,by analyzing the content of the same user's comments on different products to determine their user preferences and quantify,and the content of different users' comments on the same product to determine their product attributes and Quantify,get the deep nonlinear feature vectors of users and products,and merge this part of information with the latent hidden vectors of users and products obtained by the traditional hidden semantic model based on scoring matrix,and propose a mixed recommendation based on scoring matrix and comment text The algorithm(Hybrid recommendation algorithm based on Rating matrix and Review text,HRA-MR)avoids the problems of traditional recommendation methods and improves the recommendation effect.This article first introduces the research background and significance of the recommendation system,as well as the current status of the research on the application of recommendation algorithms and deep learning algorithms in personalized recommendation,then introduces the basic theory and related technologies required in this article,and then introduces the HRA-MR model.In this model,the review data is first grouped to obtain user review sets and product review sets,using text vectorization technology to fully mine the contextual semantic information in the reviews to obtain quantized results,and input a set of parallel convolutional neural networks.Mining user and product features;then merging the obtained user and product features with user and product hidden factor features based on the implicit semantic model;finally designing a layer of coupling structure,using two sets of features as input using factorization machine and deep neural network The combination module predicts the user's rating of the product,gets the recommendation result,and finally summarizes the conclusion,and puts forward the prospect of future research.This paper conducts experiments on 5 sets of public data sets,compares performance with a variety of classic and current advanced recommendation algorithms,and uses root mean square error(RMSE)and mean absolute error(MAE)as evaluation indicators.Experimental results show that the algorithm proposed in this paper improves the recommendation performance to a certain extent,and the prediction effect is better than that of the independent algorithms,and the score prediction accuracy has been significantly improved to varying degrees.