Research On The Parallel Algorithm Of Personalized Recommendation And Its Application

With the rapid development of Internet technology,the information on the network has become complex and diverse.Faced with massive amounts of information,it is difficult for users to obtain the content they need from it.The recommendation system is an effective information filtering technology that can solve the problem of information overload in the Internet era.In addition,the explosive growth of information has brought new problems to the recommendation system.In the face of massive information,the traditional serial recommendation algorithm is no longer suitable for today's data scale.In order to be able to mine from larger data faster to produce valuable information and design a parallel recommendation algorithm is an effective way to solve the problem of large amount of data and large amount of recommendation.This paper uses MapReduce excellent processing and analysis capabilities when faced with large data,and studies parallel algorithms for recommending large-scale data sets,and applies it to personalized takeout recommendations.The work of this article mainly has the following aspects:First,a parallel algorithm for collaborative filtering recommendation based on items,PTR-NBICF algorithm,is designed.The algorithm mainly includes two stages of data preprocessing and generating recommendation lists.The data preprocessing part crawls the review data of real products on the network.The data includes rating values and review texts.A naive Bayesian classification algorithm is used to construct sentiment classifiers for item review texts,to quantify review text sentiment values,and to combine rating values to build a comprehensive rating model.Comprehensive ratings are more capable of expressing reviewers' emotional tendencies.In the part of generating recommendation lists,the lack of similarity of items was improved,and the PTR-NBICF algorithm was implemented on the MapReduce distributed computing model,which was applied to personalized recommendation of takeout.Experiments show that the PTR-NBICF algorithm effectively improves the accuracy of item recommendation and has a good parallel acceleration ratio.Secondly,a parallel filtering recommendation algorithm based on matrix decomposition—PTR-NBALS algorithm is designed.This algorithm uses a comprehensive scoring model in the data preprocessing part,and introduces item categories into the loss function according to the principles and deficiencies of the ALS algorithm during the recommendation generation phase.Similarity is implemented on the MapReduce distributed computing model.It is used in personalized recommendations for takeout,and through in-depth mining of the takeout review data set to achieve personalized recommendations for takeout.Experiments show that the PTR-NBALS algorithm further improves the recommendation effect and enhances the reliability of the system.Finally,in view of the lack of diversity in the collaborative filtering algorithm recommendation list,based on the above research,a hybrid collaborative filtering diversified recommendation parallel algorithm-PTR-Hybrid algorithm was designed.First,the PTR-NBICF algorithm and PTR-NBALS were used The algorithm generates prediction scores,and then initially generates item recommendation candidate sets,and then inputs the candidate sets into the trained XGBoost classification model for prediction,and finally,Top-50 is retained as the final recommendation list.Experiments show that the algorithm not only has high accuracy and diversity,but also has a good acceleration ratio.It is feasible and effective to apply this algorithm to personalized and diversified takeaway recommendations.