In this paper, leisure tourism is taken as the entry point of the research, and the fused location key point features are added and integrated with the multidimensional features of time, location and space to construct an accurate portrait of social media tourism users. On the basis of tourism user profiles, a two-step clustering algorithm is combined to carry out cross-cultural analysis of social media data, to explore and excavate the performance of users’ tourism preferences under the cross-cultural ability of social media. Meanwhile, in order to realize the prediction of leisure tourism preference, a combined model based on BP neural network and ARIMA is proposed to improve the accuracy of leisure tourism preference prediction by fully considering the linear and nonlinear laws of tourism statistics. The ARIMA-BP combination prediction model is applied to predict the leisure tourism preference in the future from 2027-2034. During the period 2027-2029, the number of leisure tourism tourists maintains a high annual growth rate of more than 15%, while the growth rate slows down after 2029, with an average annual growth rate of 4.44%. In 2033, the number of leisure tourism tourists will reach 1,691,280,000, and the leisure tourism preference of tourism users has been significantly strengthened.