Research On Data-Driven Intra-City Human Mobility

Quantitative understanding of human movement behaviors via the big digital footprint data,which would provide helpful insights into human behavior patterns and urban dynamics,has become an important topic in the research of computational social science.Recently,a number of results appear in top journals like Nature and Science.However,while studying intra-city human mobility,most of existing works only explore human trajectory data in a specific city.In this thesis,we explore intra-city human mobility using trajectory data collected from five large cities in two countries.We investigate the following questions: Can we find universal patterns of human mobility in different cities? What is the mechanisms behind observations? What is the relationship between human mobility and city structure? The contributions and innovations are summarized as follows:(1)We discover several universal patterns of human mobility by taxi across cities based on the large scale taxi trajectory data from five cities that contains over 1.6 billion trips with about 30,000 taxis.First,the displacement distributions of human travel by taxi tend to follow exponential laws in two ranges where displacement larger than 2km.Second,the travel time distributions have the log-normal form.Third,for the travel time within 40 minutes,the average displacement increases linearly with travel time in the four Chinese cities.Last,the distributions of time interval between two consecutive services by the same taxi exhibit heavy-tailed characteristics,suggesting similar bursty behavior of the taxi requirements of human movements.These results indicate that there may be common mechanisms governing the behavior of human travel by taxi across cities.(2)We propose a model based on location popularity to predict intra-city population movements.The model takes into account the effects of intervening opportunities and the decision history of human movements.Theoretical result shows that in the case of uniformly distributed population and location popularity,the proposed model is equivalent to the classical gravity model.Experimental results show that it can generate displacement distribution similar to empirical data,and demonstrate its performance.(3)We introduce a combined model to predict individual locations.Based on the Markov assumption of human movement and the transfer probability among areas,a generative framework is built to describe the regularity of individual mobility.Meanwhile,matrix factorization is introduced to capture the impact of social factors on individual movement patterns.Experimental results show that the model is effective and efficient.(4)We explore the city structure of Tianjin using taxi trajectory data.From two prospectives,namely the similarity of spatio-temporal patterns,and the interaction strength among areas,we discover the two center structure and the hierarchical structure with empirical data.