Research On Tourist Behavior Data Ingestion And Intelligent Recommendation Methods

As the "Internet +" continues to empower the tourism industry,and machine learning and big data provide power for the development of the tourism industry from the software level,it has brought profound changes to the traditional tourism field.In addition to traditional group tours,Personalized travel is increasingly becoming people's travel choice.At the same time,online travel integrates tourist attractions tickets,hotels,restaurants,and entertainment resources in the online sales model,which generates a large amount of tourist behavior data.How to accurately collect and obtain these data,and then capture user preferences in a fine-grained manner has become an urgent problem to be solved.Since most of the existing sequence recommendation algorithms focus on extracting a single vector from the entire interactive sequence to represent user preferences,it is difficult to make full use of other auxiliary information.In this paper,first,aiming at the particularity of the online travel field,select the self-attention mechanism and the sequence recommendation method based on the taxonomy information to achieve high-performance recommendation on the travel e-commerce platform.At the same time,with the rapid increase of tourist behavior data,when constructing recommendation services,the training of recommendation models can hardly meet the needs of data volume and timeliness in the stand-alone mode.In order to solve the above problems,a distributed streaming processing model is introduced,and tourist behavior data is modeled as a time series data model.A distributed streaming recommendation model based on the Spark ecosystem is proposed,which uses variational inference combined with hidden factor models to produce goods recommend.The main work and contributions of this paper are as follows:(1)In order to realize the collection of tourist behavior data,we realized the access service of tourist behavior data based on Flume+Kafka+Spark Streaming,and built a tourist behavior big data platform,using visualization technology to realize the big data analysis and monitoring of the tourism industry.At the same time,on the basis of full analysis,sorting,and research,after desensitization processing,effective fields are extracted,tourist behavior characteristics are analyzed,and a tourism recommendation data set based on tourist behavior is constructed.(2)A sequence recommendation algorithm based on self-attention mechanism and product classification information for the tourism e-commerce field is proposed: SATMSRec(Self-attention based Multi-hop Sequence Recommendation for Tourism E-commerce),STMSRec fully considers the time interval between user interaction sequences And the absolute position of the sequence uses the self-attention mechanism to perform feature processing on the input sequence,and then the sequence is input into the GRU network to learn global preferences,and then combined with the product hierarchical classification information to build a multi-hop inference model to learn multi-hop preferences,in order to achieve multi-level users Preference to capture.(3)In order to solve the training task of massive data and adapt to the tourism streaming scene at the same time,Spark and Temporal Variational Inference based Distributed Tourism Streaming Recommender: TDTSR is proposed.Use Mongo DB and HDFS as distributed data storage to perform streaming input processing of data through Spark RDD.The streaming recommendation module combines the deep factorization model under the deep Bayesian learning paradigm and uses the GRU neural network combined with variational inference to construct the prediction process.Finally,Run the model under Spark On Tensorflow to achieve parallel training.