Research On Personalized Recommendation System Based On Text Mining

The rapid development of the Internet has changed people’s lifestyles,and the Internet has provided convenience for people’s lives.At the same time,it also needs to screen a huge amount of information.The recommendation system is particularly important.Various online consumer websites generate a lot of product information and review information.If you can obtain valuable content from a huge amount of textual information,you can greatly enhance the consumer’s shopping experience and promote the commodity turnover rate.This article uses the content-based recommendation idea,in addition to the description attribute of the product,mainly uses the review text,proposes a text mining-based recommendation algorithm,and improves the accuracy of the recommendation.Some commonly used recommendation algorithms are mainly cluster analysis user rating matrix,get similar user groups of target users,or recommend similar items according to user history scoring behavior.It can be seen that these algorithms rely too much on user rating information,and all users do not measure the scale,and the recommendation is not accurate.Therefore,this paper proposes a personalized recommendation algorithm based on text mining,which mainly adopts the idea of content recommendation-based algorithm.The review text is a direct description of the user’s point of view.The article product description is used to describe the corpus and the word2 vec word vector model is used to comment.Text sentiment analysis,access to product praise rate,and the selection of product positive review text set,as a description of the product’s positive feature set text,and use the LDA theme model to reduce the dimension of the review text set,extract the text feature set of the review text,calculate The article comment sets are distributed matrix of each topic.However,each topic contains many characteristic words.The use of the subject coarse-grained description comment set makes the recommendation inaccurate.The term frequency feature selection algorithm is used to select key subdivision keywords of the subject.Use topic distribution,feature word weights,and commentary rate as joint product descriptions.Combine user interest preferences with personalized recommendations.In addition to user purchase records,user ratings,product review text,and other information,items can be used when the item is cold-started.Thecontent of the intrinsic attribute is recommended to solve the cold start problem,and with the number of comments Increased gradually smooth transition to the recommendation algorithm based on text mining.The experimental part of the article crawls the watercress net movie review as experimental data for experimental analysis,using the recommended accuracy rate,recall rate as the test evaluation index.Firstly,the experiment compares the performance of the sentiment analysis before and after the recommendation system,and compares it with the traditional content-based recommendation method and the algorithm using the LDA theme recommendation directly.This paper proposes a text-based recommendation method based on the LDA-based extension of the subject text.Information and extended the LDA theme model to increase the power of text description and make the recommendation more accurate.