| Due to the irrational urban road planning and the dramatic increase of private cars in recent years,the urban traffic congestion is becoming more and more serious.In such a trend,people are turning to choose bus as the main mean of transportation.Therefore,it is very important for people to get the real-time running status of the bus and the arrival information.The bus query APPs provide the accuracy and real-time of public transport query results,which bring convenience to people and are widely used.Any operation of the users on bus query APPs will generate network log data,and these user’s operations will generate massive log data.Valuable information can be obtained from these log data.Through mining the browsing log data of APP users,service providers can find out the users’ access regular patterns and characteristics to provide users with personalized service,thus the providers could maximize commercial interests.For users themselves,after analyzing the user’s behavior data by the providers,they can get the personalized information and make their requirements be satisfied,so it can greatly reduce the time of searching information for users and brings convenience for them.With the rapid increase of user data volume,this thesis studies the user data based on Spark platform.Firstly,the thesis introduces the definition and classification of user behavior,the analysis method of user behavior analysis,and the related technologies for Spark.Then,the common clustering analysis model and correlation analysis model are compared in order to determine which the clustering model will be used in this experiment.We finally choose K-Means and SOM neural network as clustering models.The thesis selects FP-Growth model as the correlation analysis model.Secondly,the optimal clustering number is obtained by calculating the sum of squared errors.Combining the above two methods,the user behavior data of mobile APP is analyzed from the perspective of time and user interests in order to discover the distribution of users in terms of time and hobbies.The results can provide decision guidance and theoretical basis for mobile APP developers and bring certain commercial benefits to the enterprise.Finally,the clustering effects of the two clustering models are compared and analyzed by calculating the contour coefficient,and then the optimal clustering model is obtained. |
