Design And Implementation Of Personalized Tourism Recommendation System Based On Spark

Nowadays,With the rapid development of today's technology,the amount of data is also increasing day by day,people feel more and more helpless in the face of massive data.In order to solve the problem of information overload,recommendation engine came into being.Tourist market is different from e-commerce.The price index of transportation and accommodation varies greatly every day.The flow of people in different scenic spots varies greatly in different seasons.Faced with vast and changeable information,it is difficult for tourists to make decisions when making tourism plans.However,in today's Internet plus big data era,technology supporting smart tourism is gradually maturing and improving,and the policy environment is becoming more and more optimized,and the era of intelligent tourism has arrived.In this paper,firstly,the background of the argument is introduced,and the data processing technologies such as distributed computing framework Spark,file system HDFS and determinant file type Parquet are discussed in depth.Then,some algorithms and applications are introduced.On this basis,the recommendation engine is involved and implemented.The whole system is based on Spark computing framework.It uses modular partition,calls and communicates with each other to achieve decoupling.According to the particularity of online tourism,it extracts the characteristics of tourism recommendation from user groups and tourism products.The author independently establishes and implements three modules: data warehouse,recommendation engine group and result processing:(1)Data warehouse module: An efficient data warehouse is designed and implemented for off-line calculation results of recommendation engine and storage warehouse of raw data.This warehouse is based on HDFS file system.The storage file selects the column file type Parquet,which improves the efficiency of online and offline computing effectively.(2)Recommendation Engine Group: According to Spark programming model,three recommendation algorithms are parallelized.Three corresponding independent recommendation engines are designed and implemented for SVD,demography and content-based recommendation algorithms.Each engine is a small module,which works independently and produces its own recommendation results for downstream use.They can be seamlessly integrated with the underlying data warehouse,and the offline and online computing time of each engine are greatly reduced.(3)Result Processing Module: This module will produce the final recommendation results.In order to make the recommendation results of each recommendation engine consistent,a hybrid recommendation model is finally implemented.It can automatically adjust the weight of the engine(according to the user's choice),and make the personalization more fully reflected.In addition,the module also adds the function of recommendation interpretation,annotates the reason of each recommendation result and shows it to the user,and increases the user's interest and understanding.Coordination among the three modules will produce better recommendation results to users,which will cost less time and achieve better results.At present,the recommendation system has been put on line.Through continuous operation and observation,the selection of configuration parameters,user attributes and product attributes will be further adjusted.Through the operation of this system,it will be convenient for users and bring ideal economic benefits to the company.