Potential High Value Passengers Discovery And Segmentation Research Based On Random Walk

In recent years, more and more airlines began to implement customer relationship management(CRM). However, there are large numbers of potential high-value passengers who take plane fewer but have large demand of flight in future were ignored by airlines. In fact, mining these potential high-value passengers not only for airlines' potentially profitable,but also can make passengers get better service.To solve the problem that the poor of potential high-value passengers' booking data and it's hard to use the traditional method, this paper proposed to building a passengers-routes bipartite graph model to describe the relationship between the passengers and the routes based on the characteristic that bipartite graph is good at describe the contact between two different things, then solve the problem that the lack of ticket data of potential passengers by random walk on the passengers-routes bipartite graph model. The potential flight demand of passengers can be calculated by a random walk and thus calculate the potential value of passengers by combining the value of passengers and routes.After the discovery of potential high-value passengers, this paper carries out research on the subdivision of potential high-value passengers to find the different preferences of passengers, further for the airlines' personalized recommendation service support. To solve the problem that the segmentation index and the result of segmentation is rough when we segment the potential high-value passengers, this paper proposed a TCSDG model based on RFM model and the features of ticket data was be proposed in this paper which can describe the behavioral biases of passengers. It can reflect the passengers' travel time preferences,passenger cabin seat preference, discount preference and group travel preference.To solve the problem that the booking data is too huge to store and compute on the single PC when segment the potential high-value passengers according to their TCSDG value. In this paper, we parallel the subdivision algorithm which based on TCSDG model using the Hadoop parallel computing platform. Using the parallel computing platform of one host node and four computing nodes to store and analyze the ticket data, it can divided the potential high-value passengers into different clusters by their behavioral biases and provides thefoundation of personalized service and big data mining large for airlines.