School of
Information Technology and Electrical Engineering

Speaker: Dan He
Seminar Date: Fri, 27/10/2017 - 14:00
Venue: 78-631; DKE MMLab
Host: Prof Xiaofang Zhou

Seminar Type:  PhD Confirmation Seminar

Abstract: 

Trajectories between the same origin and destination (OD) offer valuable information for us to better understand the diversity of moving behaviours and the intrinsic relationships between the moving objects and specific locations, which foster plenty of applications in location-based social networks and intelligent transportation services, e.g., the personalized routing service. In particular, given an origin o and a destination d, existing online navigation services, e.g., Google Map, usually provide the shortest path or the fastest route from o to d. Nevertheless, there are many cases that some routes frequently traversed by local drivers, are more reasonable choices but are never recommended by the online navigation service. The personalized routing service, which recommends users routes by mining the historical trajectory data, can easily find such routes and make recommendations. Through the analytics on historical trajectory data, one can provide more options for users according to their personal preferences, e.g., the fastest path, the quite roads, or the most popular route.  Consequently, in order to support the personalized routing, some desirable features are  able to identify different categories of historical trajectories and assign meaningful labels to the categories based on their characteristics. However, such task turns out to be quite difficult due to the sparsity issue that exists in trajectory datasets. To explain, no matter how large the trajectory dataset might be, the number of trajectories might still be small under certain constraints, e.g. across a given origin-destination pair of points/regions within a specific time duration. Motivated by this, the first and fundamental problem that we aim to address is the sparsity issue, i.e., we need to get sizable trajectories to represent the moving behaviours from an origin to a destination. Afterwards, in order to provide better understanding on moving behaviours w.r.t. an OD pair, we further perform clustering on the corresponding trajectories, followed by the labelling for different categories of trajectories based on their intrinsic features. The current results of our empirical study demonstrate the efficiency and effectiveness of our trajectory augmentation approach to address the trajectory sparsity issue.

Biography: 

Ms. Dan He obtained her Bachelor of Science (Information Security) in 2012 from the University of Science & Technology Beijing, China. Later she obtained Master of Science (Computer Science) from the Peking University, Beijing, China in 2015. Currently she is a PhD student at DKE group under supervision of Prof. Xiaofang Zhou (principal), and Dr. Sibo Wang (associate). Her research interests include spatial-temporal data management and high performance query processing.