Research And Development Of Personalized Travel Recommendation System Based On Flink

As the tertiary industry of our country,tourism is an important part of the national economy.With the development of 5G,cloud computing,big data,Internet of things,artificial intelligence and other technologies,the explosive growth of data makes it difficult for people to find travel information that meets their needs from massive data,but big data technology can process massive data and personalize travel.The recommendation system can actively recommend based on relevant data,which can greatly alleviate this problem.The application scenarios and environments of the current travel recommendation system are complex and changeable,and related research and development has huge space and value.Therefore,a personalized travel recommendation system based on Flink is researched and developed in this dissertation.The main work is as follows:1.Based on the new generation of big data computing engine Flink to implement the system analysis and computing layer,the implementation of offline and real-time recommendation algorithms is compared with the current popular big data computing engine Spark.The results show that the implementation based on Flink achieves lower recommendation delay.2.In the offline recommendation module,the collaborative filtering algorithm based on the hidden semantic model is used to alleviate the sparsity of the scoring matrix through matrix decomposition,and the model is solved based on the alternating least squares(ALS)algorithm.To address the problem that the collaborative filtering model cannot use more information,a recommendation model based on ALS-based collaborative filtering and extreme gradient boosting tree(XgBoost)weighted fusion is proposed.Experiments show that the model has a lower recommendation prediction error.3.In the real-time recommendation module,the basic similarity list of scenic spots is formed according to the latent features of the scenic spots in the rating matrix.Considering the user rating preference factor,the similarity of scenic spots and the recent ratings of users are weighted,and a reward and punishment function is proposed to form the final recommendation calculation results.The effective impact of this function on the recommendation results is verified,and completes the real-time update recommendation of recent popular and most rated modules.4.The Qunar.com’s data about all scenic spots in Guangxi are crawled using Web crawler technology,and then a set of Guangxi personalized tourism big data recommendation system is developed based on Kafka,Redis,Flume,Elasticsearch,front-end and back-end,distributed and other technologies.The main functions of the system include similar recommendations based on scenic spot content,offline recommendations for scenic spots,real-time recommendations for scenic spots,recent popular recommendations,recommendations with the highest ratings,and recommendations for graded scenic spots,which show that the system achieves personalization and diversification of recommendations.